HP schaltet ohne Vorankündigung HP ePrint ab

HP schaltet ohne Vorankündigung HP ePrint ab

Bedauerlicherweise muss ich dir mitteilen, es ist normal, dass du dich mit deinem HP Photosmart 7515 bei HP eprint / HP Connected oder über die Webdienste nicht mehr anmelden kannst.

Auf Grund des Alters dieses Druckermodells wurde E-Print im September 2016 für diesen Drucker eingestellt.
Der Drucker hatte bereits Ende 2015 das Enddatum seines Supports erreicht.

Es gibt keine Möglichkeit mehr, mit deinem 7515 die eprint-Funktion zu nutzen.
Tut mir Leid.

theatlantic.com: My Family’s Slave

theatlantic.com: My Family’s Slave

The ashes filled a black plastic box about the size of a toaster. It weighed three and a half pounds. I put it in a canvas tote bag and packed it in my suitcase this past July for the transpacific flight to Manila. From there I would travel by car to a rural village. When I arrived, I would hand over all that was left of the woman who had spent 56 years as a slave in my family’s household.

Her name was Eudocia Tomas Pulido. We called her Lola. She was 4 foot 11, with mocha-brown skin and almond eyes that I can still see looking into mine—my first memory. She was 18 years old when my grandfather gave her to my mother as a gift, and when my family moved to the United States, we brought her with us. No other word but slave encompassed the life she lived. Her days began before everyone else woke and ended after we went to bed. She prepared three meals a day, cleaned the house, waited on my parents, and took care of my four siblings and me. My parents never paid her, and they scolded her constantly. She wasn’t kept in leg irons, but she might as well have been. So many nights, on my way to the bathroom, I’d spot her sleeping in a corner, slumped against a mound of laundry, her fingers clutching a garment she was in the middle of folding.

To our American neighbors, we were model immigrants, a poster family. They told us so. My father had a law degree, my mother was on her way to becoming a doctor, and my siblings and I got good grades and always said “please” and “thank you.” We never talked about Lola. Our secret went to the core of who we were and, at least for us kids, who we wanted to be.

After my mother died of leukemia, in 1999, Lola came to live with me in a small town north of Seattle. I had a family, a career, a house in the suburbs—the American dream. And then I had a slave.

At baggage claim in Manila, I unzipped my suitcase to make sure Lola’s ashes were still there. Outside, I inhaled the familiar smell: a thick blend of exhaust and waste, of ocean and sweet fruit and sweat.

Early the next morning I found a driver, an affable middle-aged man who went by the nickname “Doods,” and we hit the road in his truck, weaving through traffic. The scene always stunned me. The sheer number of cars and motorcycles and jeepneys. The people weaving between them and moving on the sidewalks in great brown rivers. The street vendors in bare feet trotting alongside cars, hawking cigarettes and cough drops and sacks of boiled peanuts. The child beggars pressing their faces against the windows.

Doods and I were headed to the place where Lola’s story began, up north in the central plains: Tarlac province. Rice country. The home of a cigar-chomping army lieutenant named Tomas Asuncion, my grandfather. The family stories paint Lieutenant Tom as a formidable man given to eccentricity and dark moods, who had lots of land but little money and kept mistresses in separate houses on his property. His wife died giving birth to their only child, my mother. She was raised by a series of utusans, or “people who take commands.”

Slavery has a long history on the islands. Before the Spanish came, islanders enslaved other islanders, usually war captives, criminals, or debtors. Slaves came in different varieties, from warriors who could earn their freedom through valor to household servants who were regarded as property and could be bought and sold or traded. High-status slaves could own low-status slaves, and the low could own the lowliest. Some chose to enter servitude simply to survive: In exchange for their labor, they might be given food, shelter, and protection.

When the Spanish arrived, in the 1500s, they enslaved islanders and later brought African and Indian slaves. The Spanish Crown eventually began phasing out slavery at home and in its colonies, but parts of the Philippines were so far-flung that authorities couldn’t keep a close eye. Traditions persisted under different guises, even after the U.S. took control of the islands in 1898. Today even the poor can have utusans or katulongs (“helpers”) or kasambahays (“domestics”), as long as there are people even poorer. The pool is deep.

Lieutenant Tom had as many as three families of utusans living on his property. In the spring of 1943, with the islands under Japanese occupation, he brought home a girl from a village down the road. She was a cousin from a marginal side of the family, rice farmers. The lieutenant was shrewd—he saw that this girl was penniless, unschooled, and likely to be malleable. Her parents wanted her to marry a pig farmer twice her age, and she was desperately unhappy but had nowhere to go. Tom approached her with an offer: She could have food and shelter if she would commit to taking care of his daughter, who had just turned 12.

Lola agreed, not grasping that the deal was for life.

“She is my gift to you,” Lieutenant Tom told my mother.

“I don’t want her,” my mother said, knowing she had no choice.

Lieutenant Tom went off to fight the Japanese, leaving Mom behind with Lola in his creaky house in the provinces. Lola fed, groomed, and dressed my mother. When they walked to the market, Lola held an umbrella to shield her from the sun. At night, when Lola’s other tasks were done—feeding the dogs, sweeping the floors, folding the laundry that she had washed by hand in the Camiling River—she sat at the edge of my mother’s bed and fanned her to sleep.

One day during the war Lieutenant Tom came home and caught my mother in a lie—something to do with a boy she wasn’t supposed to talk to. Tom, furious, ordered her to “stand at the table.” Mom cowered with Lola in a corner. Then, in a quivering voice, she told her father that Lola would take her punishment. Lola looked at Mom pleadingly, then without a word walked to the dining table and held on to the edge. Tom raised the belt and delivered 12 lashes, punctuating each one with a word. You. Do. Not. Lie. To. Me. You. Do. Not. Lie. To. Me. Lola made no sound.

My mother, in recounting this story late in her life, delighted in the outrageousness of it, her tone seeming to say, Can you believe I did that? When I brought it up with Lola, she asked to hear Mom’s version. She listened intently, eyes lowered, and afterward she looked at me with sadness and said simply, “Yes. It was like that.”

Seven years later, in 1950, Mom married my father and moved to Manila, bringing Lola along. Lieutenant Tom had long been haunted by demons, and in 1951 he silenced them with a .32‑caliber slug to his temple. Mom almost never talked about it. She had his temperament—moody, imperial, secretly fragile—and she took his lessons to heart, among them the proper way to be a provincial matrona: You must embrace your role as the giver of commands. You must keep those beneath you in their place at all times, for their own good and the good of the household. They might cry and complain, but their souls will thank you. They will love you for helping them be what God intended.

My brother Arthur was born in 1951. I came next, followed by three more siblings in rapid succession. My parents expected Lola to be as devoted to us kids as she was to them. While she looked after us, my parents went to school and earned advanced degrees, joining the ranks of so many others with fancy diplomas but no jobs. Then the big break: Dad was offered a job in Foreign Affairs as a commercial analyst. The salary would be meager, but the position was in America—a place he and Mom had grown up dreaming of, where everything they hoped for could come true.

Dad was allowed to bring his family and one domestic. Figuring they would both have to work, my parents needed Lola to care for the kids and the house. My mother informed Lola, and to her great irritation, Lola didn’t immediately acquiesce. Years later Lola told me she was terrified. “It was too far,” she said. “Maybe your Mom and Dad won’t let me go home.”

In the end what convinced Lola was my father’s promise that things would be different in America. He told her that as soon as he and Mom got on their feet, they’d give her an “allowance.” Lola could send money to her parents, to all her relations in the village. Her parents lived in a hut with a dirt floor. Lola could build them a concrete house, could change their lives forever. Imagine.

We landed in Los Angeles on May 12, 1964, all our belongings in cardboard boxes tied with rope. Lola had been with my mother for 21 years by then. In many ways she was more of a parent to me than either my mother or my father. Hers was the first face I saw in the morning and the last one I saw at night. As a baby, I uttered Lola’s name (which I first pronounced “Oh-ah”) long before I learned to say “Mom” or “Dad.” As a toddler, I refused to go to sleep unless Lola was holding me, or at least nearby.

I was 4 years old when we arrived in the U.S.—too young to question Lola’s place in our family. But as my siblings and I grew up on this other shore, we came to see the world differently. The leap across the ocean brought about a leap in consciousness that Mom and Dad couldn’t, or wouldn’t, make.

Lola never got that allowance. She asked my parents about it in a roundabout way a couple of years into our life in America. Her mother had fallen ill (with what I would later learn was dysentery), and her family couldn’t afford the medicine she needed. “Pwede ba?” she said to my parents. Is it possible? Mom let out a sigh. “How could you even ask?,” Dad responded in Tagalog. “You see how hard up we are. Don’t you have any shame?”

My parents had borrowed money for the move to the U.S., and then borrowed more in order to stay. My father was transferred from the consulate general in L.A. to the Philippine consulate in Seattle. He was paid $5,600 a year. He took a second job cleaning trailers, and a third as a debt collector. Mom got work as a technician in a couple of medical labs. We barely saw them, and when we did they were often exhausted and snappish.

Mom would come home and upbraid Lola for not cleaning the house well enough or for forgetting to bring in the mail. “Didn’t I tell you I want the letters here when I come home?” she would say in Tagalog, her voice venomous. “It’s not hard naman! An idiot could remember.” Then my father would arrive and take his turn. When Dad raised his voice, everyone in the house shrank. Sometimes my parents would team up until Lola broke down crying, almost as though that was their goal.

It confused me: My parents were good to my siblings and me, and we loved them. But they’d be affectionate to us kids one moment and vile to Lola the next. I was 11 or 12 when I began to see Lola’s situation clearly. By then Arthur, eight years my senior, had been seething for a long time. He was the one who introduced the word slave into my understanding of what Lola was. Before he said it I’d thought of her as just an unfortunate member of the household. I hated when my parents yelled at her, but it hadn’t occurred to me that they—and the whole arrangement—could be immoral.

“Do you know anybody treated the way she’s treated?,” Arthur said. “Who lives the way she lives?” He summed up Lola’s reality: Wasn’t paid. Toiled every day. Was tongue-lashed for sitting too long or falling asleep too early. Was struck for talking back. Wore hand-me-downs. Ate scraps and leftovers by herself in the kitchen. Rarely left the house. Had no friends or hobbies outside the family. Had no private quarters. (Her designated place to sleep in each house we lived in was always whatever was left—a couch or storage area or corner in my sisters’ bedroom. She often slept among piles of laundry.)

We couldn’t identify a parallel anywhere except in slave characters on TV and in the movies. I remember watching a Western called The Man Who Shot Liberty Valance. John Wayne plays Tom Doniphon, a gunslinging rancher who barks orders at his servant, Pompey, whom he calls his “boy.” Pick him up, Pompey. Pompey, go find the doctor. Get on back to work, Pompey! Docile and obedient, Pompey calls his master “Mistah Tom.” They have a complex relationship. Tom forbids Pompey from attending school but opens the way for Pompey to drink in a whites-only saloon. Near the end, Pompey saves his master from a fire. It’s clear Pompey both fears and loves Tom, and he mourns when Tom dies. All of this is peripheral to the main story of Tom’s showdown with bad guy Liberty Valance, but I couldn’t take my eyes off Pompey. I remember thinking: Lola is Pompey, Pompey is Lola.

One night when Dad found out that my sister Ling, who was then 9, had missed dinner, he barked at Lola for being lazy. “I tried to feed her,” Lola said, as Dad stood over her and glared. Her feeble defense only made him angrier, and he punched her just below the shoulder. Lola ran out of the room and I could hear her wailing, an animal cry.

“Ling said she wasn’t hungry,” I said.

My parents turned to look at me. They seemed startled. I felt the twitching in my face that usually preceded tears, but I wouldn’t cry this time. In Mom’s eyes was a shadow of something I hadn’t seen before. Jealousy?

“Are you defending your Lola?,” Dad said. “Is that what you’re doing?”

“Ling said she wasn’t hungry,” I said again, almost in a whisper.

I was 13. It was my first attempt to stick up for the woman who spent her days watching over me. The woman who used to hum Tagalog melodies as she rocked me to sleep, and when I got older would dress and feed me and walk me to school in the mornings and pick me up in the afternoons. Once, when I was sick for a long time and too weak to eat, she chewed my food for me and put the small pieces in my mouth to swallow. One summer when I had plaster casts on both legs (I had problem joints), she bathed me with a washcloth, brought medicine in the middle of the night, and helped me through months of rehabilitation. I was cranky through it all. She didn’t complain or lose patience, ever.

To now hear her wailing made me crazy.

In the old country, my parents felt no need to hide their treatment of Lola. In America, they treated her worse but took pains to conceal it. When guests came over, my parents would either ignore her or, if questioned, lie and quickly change the subject. For five years in North Seattle, we lived across the street from the Misslers, a rambunctious family of eight who introduced us to things like mustard, salmon fishing, and mowing the lawn. Football on TV. Yelling during football. Lola would come out to serve food and drinks during games, and my parents would smile and thank her before she quickly disappeared. “Who’s that little lady you keep in the kitchen?,” Big Jim, the Missler patriarch, once asked. A relative from back home, Dad said. Very shy.

Billy Missler, my best friend, didn’t buy it. He spent enough time at our house, whole weekends sometimes, to catch glimpses of my family’s secret. He once overheard my mother yelling in the kitchen, and when he barged in to investigate found Mom red-faced and glaring at Lola, who was quaking in a corner. I came in a few seconds later. The look on Billy’s face was a mix of embarrassment and perplexity. What was that? I waved it off and told him to forget it.

I think Billy felt sorry for Lola. He’d rave about her cooking, and make her laugh like I’d never seen. During sleepovers, she’d make his favorite Filipino dish, beef tapa over white rice. Cooking was Lola’s only eloquence. I could tell by what she served whether she was merely feeding us or saying she loved us.

When I once referred to Lola as a distant aunt, Billy reminded me that when we’d first met I’d said she was my grandmother.

“Well, she’s kind of both,” I said mysteriously.

“Why is she always working?”

“She likes to work,” I said.

“Your dad and mom—why do they yell at her?”

“Her hearing isn’t so good …”

Admitting the truth would have meant exposing us all. We spent our first decade in the country learning the ways of the new land and trying to fit in. Having a slave did not fit. Having a slave gave me grave doubts about what kind of people we were, what kind of place we came from. Whether we deserved to be accepted. I was ashamed of it all, including my complicity. Didn’t I eat the food she cooked, and wear the clothes she washed and ironed and hung in the closet? But losing her would have been devastating.

There was another reason for secrecy: Lola’s travel papers had expired in 1969, five years after we arrived in the U.S. She’d come on a special passport linked to my father’s job. After a series of fallings-out with his superiors, Dad quit the consulate and declared his intent to stay in the United States. He arranged for permanent-resident status for his family, but Lola wasn’t eligible. He was supposed to send her back.

Lola’s mother, Fermina, died in 1973; her father, Hilario, in 1979. Both times she wanted desperately to go home. Both times my parents said “Sorry.” No money, no time. The kids needed her. My parents also feared for themselves, they admitted to me later. If the authorities had found out about Lola, as they surely would have if she’d tried to leave, my parents could have gotten into trouble, possibly even been deported. They couldn’t risk it. Lola’s legal status became what Filipinos call tago nang tago, or TNT—“on the run.” She stayed TNT for almost 20 years.

After each of her parents died, Lola was sullen and silent for months. She barely responded when my parents badgered her. But the badgering never let up. Lola kept her head down and did her work.

My father’s resignation started a turbulent period. Money got tighter, and my parents turned on each other. They uprooted the family again and again—Seattle to Honolulu back to Seattle to the southeast Bronx and finally to the truck-stop town of Umatilla, Oregon, population 750. During all this moving around, Mom often worked 24-hour shifts, first as a medical intern and then as a resident, and Dad would disappear for days, working odd jobs but also (we’d later learn) womanizing and who knows what else. Once, he came home and told us that he’d lost our new station wagon playing blackjack.

For days in a row Lola would be the only adult in the house. She got to know the details of our lives in a way that my parents never had the mental space for. We brought friends home, and she’d listen to us talk about school and girls and boys and whatever else was on our minds. Just from conversations she overheard, she could list the first name of every girl I had a crush on from sixth grade through high school.

When I was 15, Dad left the family for good. I didn’t want to believe it at the time, but the fact was that he deserted us kids and abandoned Mom after 25 years of marriage. She wouldn’t become a licensed physician for another year, and her specialty—internal medicine—wasn’t especially lucrative. Dad didn’t pay child support, so money was always a struggle.

My mom kept herself together enough to go to work, but at night she’d crumble in self-pity and despair. Her main source of comfort during this time: Lola. As Mom snapped at her over small things, Lola attended to her even more—cooking Mom’s favorite meals, cleaning her bedroom with extra care. I’d find the two of them late at night at the kitchen counter, griping and telling stories about Dad, sometimes laughing wickedly, other times working themselves into a fury over his transgressions. They barely noticed us kids flitting in and out.

One night I heard Mom weeping and ran into the living room to find her slumped in Lola’s arms. Lola was talking softly to her, the way she used to with my siblings and me when we were young. I lingered, then went back to my room, scared for my mom and awed by Lola.

Doods was humming. I’d dozed for what felt like a minute and awoke to his happy melody. “Two hours more,” he said. I checked the plastic box in the tote bag by my side—still there—and looked up to see open road. The MacArthur Highway. I glanced at the time. “Hey, you said ‘two hours’ two hours ago,” I said. Doods just hummed.

His not knowing anything about the purpose of my journey was a relief. I had enough interior dialogue going on. I was no better than my parents. I could have done more to free Lola. To make her life better. Why didn’t I? I could have turned in my parents, I suppose. It would have blown up my family in an instant. Instead, my siblings and I kept everything to ourselves, and rather than blowing up in an instant, my family broke apart slowly.

Doods and I passed through beautiful country. Not travel-brochure beautiful but real and alive and, compared with the city, elegantly spare. Mountains ran parallel to the highway on each side, the Zambales Mountains to the west, the Sierra Madre Range to the east. From ridge to ridge, west to east, I could see every shade of green all the way to almost black.

Doods pointed to a shadowy outline in the distance. Mount Pinatubo. I’d come here in 1991 to report on the aftermath of its eruption, the second-largest of the 20th century. Volcanic mudflows called lahars continued for more than a decade, burying ancient villages, filling in rivers and valleys, and wiping out entire ecosystems. The lahars reached deep into the foothills of Tarlac province, where Lola’s parents had spent their entire lives, and where she and my mother had once lived together. So much of our family record had been lost in wars and floods, and now parts were buried under 20 feet of mud.

Life here is routinely visited by cataclysm. Killer typhoons that strike several times a year. Bandit insurgencies that never end. Somnolent mountains that one day decide to wake up. The Philippines isn’t like China or Brazil, whose mass might absorb the trauma. This is a nation of scattered rocks in the sea. When disaster hits, the place goes under for a while. Then it resurfaces and life proceeds, and you can behold a scene like the one Doods and I were driving through, and the simple fact that it’s still there makes it beautiful.

A couple of years after my parents split, my mother remarried and demanded Lola’s fealty to her new husband, a Croatian immigrant named Ivan, whom she had met through a friend. Ivan had never finished high school. He’d been married four times and was an inveterate gambler who enjoyed being supported by my mother and attended to by Lola.

Ivan brought out a side of Lola I’d never seen. His marriage to my mother was volatile from the start, and money—especially his use of her money—was the main issue. Once, during an argument in which Mom was crying and Ivan was yelling, Lola walked over and stood between them. She turned to Ivan and firmly said his name. He looked at Lola, blinked, and sat down.

My sister Inday and I were floored. Ivan was about 250 pounds, and his baritone could shake the walls. Lola put him in his place with a single word. I saw this happen a few other times, but for the most part Lola served Ivan unquestioningly, just as Mom wanted her to. I had a hard time watching Lola vassalize herself to another person, especially someone like Ivan. But what set the stage for my blowup with Mom was something more mundane.

She used to get angry whenever Lola felt ill. She didn’t want to deal with the disruption and the expense, and would accuse Lola of faking or failing to take care of herself. Mom chose the second tack when, in the late 1970s, Lola’s teeth started falling out. She’d been saying for months that her mouth hurt.

“That’s what happens when you don’t brush properly,” Mom told her.

I said that Lola needed to see a dentist. She was in her 50s and had never been to one. I was attending college an hour away, and I brought it up again and again on my frequent trips home. A year went by, then two. Lola took aspirin every day for the pain, and her teeth looked like a crumbling Stonehenge. One night, after watching her chew bread on the side of her mouth that still had a few good molars, I lost it.

Mom and I argued into the night, each of us sobbing at different points. She said she was tired of working her fingers to the bone supporting everybody, and sick of her children always taking Lola’s side, and why didn’t we just take our goddamn Lola, she’d never wanted her in the first place, and she wished to God she hadn’t given birth to an arrogant, sanctimonious phony like me.

I let her words sink in. Then I came back at her, saying she would know all about being a phony, her whole life was a masquerade, and if she stopped feeling sorry for herself for one minute she’d see that Lola could barely eat because her goddamn teeth were rotting out of her goddamn head, and couldn’t she think of her just this once as a real person instead of a slave kept alive to serve her?

“A slave,” Mom said, weighing the word. “A slave?”

The night ended when she declared that I would never understand her relationship with Lola. Never. Her voice was so guttural and pained that thinking of it even now, so many years later, feels like a punch to the stomach. It’s a terrible thing to hate your own mother, and that night I did. The look in her eyes made clear that she felt the same way about me.

The fight only fed Mom’s fear that Lola had stolen the kids from her, and she made Lola pay for it. Mom drove her harder. Tormented her by saying, “I hope you’re happy now that your kids hate me.” When we helped Lola with housework, Mom would fume. “You’d better go to sleep now, Lola,” she’d say sarcastically. “You’ve been working too hard. Your kids are worried about you.” Later she’d take Lola into a bedroom for a talk, and Lola would walk out with puffy eyes.

Lola finally begged us to stop trying to help her.

Why do you stay? we asked.

“Who will cook?” she said, which I took to mean, Who would do everything? Who would take care of us? Of Mom? Another time she said, “Where will I go?” This struck me as closer to a real answer. Coming to America had been a mad dash, and before we caught a breath a decade had gone by. We turned around, and a second decade was closing out. Lola’s hair had turned gray. She’d heard that relatives back home who hadn’t received the promised support were wondering what had happened to her. She was ashamed to return.

She had no contacts in America, and no facility for getting around. Phones puzzled her. Mechanical things—ATMs, intercoms, vending machines, anything with a keyboard—made her panic. Fast-talking people left her speechless, and her own broken English did the same to them. She couldn’t make an appointment, arrange a trip, fill out a form, or order a meal without help.

I got Lola an ATM card linked to my bank account and taught her how to use it. She succeeded once, but the second time she got flustered, and she never tried again. She kept the card because she considered it a gift from me.

I also tried to teach her to drive. She dismissed the idea with a wave of her hand, but I picked her up and carried her to the car and planted her in the driver’s seat, both of us laughing. I spent 20 minutes going over the controls and gauges. Her eyes went from mirthful to terrified. When I turned on the ignition and the dashboard lit up, she was out of the car and in the house before I could say another word. I tried a couple more times.

I thought driving could change her life. She could go places. And if things ever got unbearable with Mom, she could drive away forever.

Four lanes became two, pavement turned to gravel. Tricycle drivers wove between cars and water buffalo pulling loads of bamboo. An occasional dog or goat sprinted across the road in front of our truck, almost grazing the bumper. Doods never eased up. Whatever didn’t make it across would be stew today instead of tomorrow—the rule of the road in the provinces.

I took out a map and traced the route to the village of Mayantoc, our destination. Out the window, in the distance, tiny figures folded at the waist like so many bent nails. People harvesting rice, the same way they had for thousands of years. We were getting close.

I tapped the cheap plastic box and regretted not buying a real urn, made of porcelain or rosewood. What would Lola’s people think? Not that many were left. Only one sibling remained in the area, Gregoria, 98 years old, and I was told her memory was failing. Relatives said that whenever she heard Lola’s name, she’d burst out crying and then quickly forget why.

I’d been in touch with one of Lola’s nieces. She had the day planned: When I arrived, a low-key memorial, then a prayer, followed by the lowering of the ashes into a plot at the Mayantoc Eternal Bliss Memorial Park. It had been five years since Lola died, but I hadn’t yet said the final goodbye that I knew was about to happen. All day I had been feeling intense grief and resisting the urge to let it out, not wanting to wail in front of Doods. More than the shame I felt for the way my family had treated Lola, more than my anxiety about how her relatives in Mayantoc would treat me, I felt the terrible heaviness of losing her, as if she had died only the day before.

Doods veered northwest on the Romulo Highway, then took a sharp left at Camiling, the town Mom and Lieutenant Tom came from. Two lanes became one, then gravel turned to dirt. The path ran along the Camiling River, clusters of bamboo houses off to the side, green hills ahead. The homestretch.

I gave the eulogy at Mom’s funeral, and everything I said was true. That she was brave and spirited. That she’d drawn some short straws, but had done the best she could. That she was radiant when she was happy. That she adored her children, and gave us a real home—in Salem, Oregon—that through the ’80s and ’90s became the permanent base we’d never had before. That I wished we could thank her one more time. That we all loved her.

I didn’t talk about Lola. Just as I had selectively blocked Lola out of my mind when I was with Mom during her last years. Loving my mother required that kind of mental surgery. It was the only way we could be mother and son—which I wanted, especially after her health started to decline, in the mid‑’90s. Diabetes. Breast cancer. Acute myelogenous leukemia, a fast-growing cancer of the blood and bone marrow. She went from robust to frail seemingly overnight.

After the big fight, I mostly avoided going home, and at age 23 I moved to Seattle. When I did visit I saw a change. Mom was still Mom, but not as relentlessly. She got Lola a fine set of dentures and let her have her own bedroom. She cooperated when my siblings and I set out to change Lola’s TNT status. Ronald Reagan’s landmark immigration bill of 1986 made millions of illegal immigrants eligible for amnesty. It was a long process, but Lola became a citizen in October 1998, four months after my mother was diagnosed with leukemia. Mom lived another year.

During that time, she and Ivan took trips to Lincoln City, on the Oregon coast, and sometimes brought Lola along. Lola loved the ocean. On the other side were the islands she dreamed of returning to. And Lola was never happier than when Mom relaxed around her. An afternoon at the coast or just 15 minutes in the kitchen reminiscing about the old days in the province, and Lola would seem to forget years of torment.

I couldn’t forget so easily. But I did come to see Mom in a different light. Before she died, she gave me her journals, two steamer trunks’ full. Leafing through them as she slept a few feet away, I glimpsed slices of her life that I’d refused to see for years. She’d gone to medical school when not many women did. She’d come to America and fought for respect as both a woman and an immigrant physician. She’d worked for two decades at Fairview Training Center, in Salem, a state institution for the developmentally disabled. The irony: She tended to underdogs most of her professional life. They worshipped her. Female colleagues became close friends. They did silly, girly things together—shoe shopping, throwing dress-up parties at one another’s homes, exchanging gag gifts like penis-shaped soaps and calendars of half-naked men, all while laughing hysterically. Looking through their party pictures reminded me that Mom had a life and an identity apart from the family and Lola. Of course.

Mom wrote in great detail about each of her kids, and how she felt about us on a given day—proud or loving or resentful. And she devoted volumes to her husbands, trying to grasp them as complex characters in her story. We were all persons of consequence. Lola was incidental. When she was mentioned at all, she was a bit character in someone else’s story. “Lola walked my beloved Alex to his new school this morning. I hope he makes new friends quickly so he doesn’t feel so sad about moving again …” There might be two more pages about me, and no other mention of Lola.

The day before Mom died, a Catholic priest came to the house to perform last rites. Lola sat next to my mother’s bed, holding a cup with a straw, poised to raise it to Mom’s mouth. She had become extra attentive to my mother, and extra kind. She could have taken advantage of Mom in her feebleness, even exacted revenge, but she did the opposite.

The priest asked Mom whether there was anything she wanted to forgive or be forgiven for. She scanned the room with heavy-lidded eyes, said nothing. Then, without looking at Lola, she reached over and placed an open hand on her head. She didn’t say a word.

Lola was 75 when she came to stay with me. I was married with two young daughters, living in a cozy house on a wooded lot. From the second story, we could see Puget Sound. We gave Lola a bedroom and license to do whatever she wanted: sleep in, watch soaps, do nothing all day. She could relax—and be free—for the first time in her life. I should have known it wouldn’t be that simple.

I’d forgotten about all the things Lola did that drove me a little crazy. She was always telling me to put on a sweater so I wouldn’t catch a cold (I was in my 40s). She groused incessantly about Dad and Ivan: My father was lazy, Ivan was a leech. I learned to tune her out. Harder to ignore was her fanatical thriftiness. She threw nothing out. And she used to go through the trash to make sure that the rest of us hadn’t thrown out anything useful. She washed and reused paper towels again and again until they disintegrated in her hands. (No one else would go near them.) The kitchen became glutted with grocery bags, yogurt containers, and pickle jars, and parts of our house turned into storage for—there’s no other word for it—garbage.

She cooked breakfast even though none of us ate more than a banana or a granola bar in the morning, usually while we were running out the door. She made our beds and did our laundry. She cleaned the house. I found myself saying to her, nicely at first, “Lola, you don’t have to do that.” “Lola, we’ll do it ourselves.” “Lola, that’s the girls’ job.” Okay, she’d say, but keep right on doing it.

It irritated me to catch her eating meals standing in the kitchen, or see her tense up and start cleaning when I walked into the room. One day, after several months, I sat her down.

“I’m not Dad. You’re not a slave here,” I said, and went through a long list of slavelike things she’d been doing. When I realized she was startled, I took a deep breath and cupped her face, that elfin face now looking at me searchingly. I kissed her forehead. “This is your house now,” I said. “You’re not here to serve us. You can relax, okay?”

“Okay,” she said. And went back to cleaning.

She didn’t know any other way to be. I realized I had to take my own advice and relax. If she wanted to make dinner, let her. Thank her and do the dishes. I had to remind myself constantly: Let her be.

One night I came home to find her sitting on the couch doing a word puzzle, her feet up, the TV on. Next to her, a cup of tea. She glanced at me, smiled sheepishly with those perfect white dentures, and went back to the puzzle. Progress, I thought.

She planted a garden in the backyard—roses and tulips and every kind of orchid—and spent whole afternoons tending it. She took walks around the neighborhood. At about 80, her arthritis got bad and she began walking with a cane. In the kitchen she went from being a fry cook to a kind of artisanal chef who created only when the spirit moved her. She made lavish meals and grinned with pleasure as we devoured them.

Passing the door of Lola’s bedroom, I’d often hear her listening to a cassette of Filipino folk songs. The same tape over and over. I knew she’d been sending almost all her money—my wife and I gave her $200 a week—to relatives back home. One afternoon, I found her sitting on the back deck gazing at a snapshot someone had sent of her village.

“You want to go home, Lola?”

She turned the photograph over and traced her finger across the inscription, then flipped it back and seemed to study a single detail.

“Yes,” she said.

Just after her 83rd birthday, I paid her airfare to go home. I’d follow a month later to bring her back to the U.S.—if she wanted to return. The unspoken purpose of her trip was to see whether the place she had spent so many years longing for could still feel like home.

She found her answer.

“Everything was not the same,” she told me as we walked around Mayantoc. The old farms were gone. Her house was gone. Her parents and most of her siblings were gone. Childhood friends, the ones still alive, were like strangers. It was nice to see them, but … everything was not the same. She’d still like to spend her last years here, she said, but she wasn’t ready yet.

“You’re ready to go back to your garden,” I said.

“Yes. Let’s go home.”

Lola was as devoted to my daughters as she’d been to my siblings and me when we were young. After school, she’d listen to their stories and make them something to eat. And unlike my wife and me (especially me), Lola enjoyed every minute of every school event and performance. She couldn’t get enough of them. She sat up front, kept the programs as mementos.

It was so easy to make Lola happy. We took her on family vacations, but she was as excited to go to the farmer’s market down the hill. She became a wide-eyed kid on a field trip: “Look at those zucchinis!” The first thing she did every morning was open all the blinds in the house, and at each window she’d pause to look outside.

And she taught herself to read. It was remarkable. Over the years, she’d somehow learned to sound out letters. She did those puzzles where you find and circle words within a block of letters. Her room had stacks of word-puzzle booklets, thousands of words circled in pencil. Every day she watched the news and listened for words she recognized. She triangulated them with words in the newspaper, and figured out the meanings. She came to read the paper every day, front to back. Dad used to say she was simple. I wondered what she could have been if, instead of working the rice fields at age 8, she had learned to read and write.

During the 12 years she lived in our house, I asked her questions about herself, trying to piece together her life story, a habit she found curious. To my inquiries she would often respond first with “Why?” Why did I want to know about her childhood? About how she met Lieutenant Tom?

I tried to get my sister Ling to ask Lola about her love life, thinking Lola would be more comfortable with her. Ling cackled, which was her way of saying I was on my own. One day, while Lola and I were putting away groceries, I just blurted it out: “Lola, have you ever been romantic with anyone?” She smiled, and then she told me the story of the only time she’d come close. She was about 15, and there was a handsome boy named Pedro from a nearby farm. For several months they harvested rice together side by side. One time, she dropped her bolo—a cutting implement—and he quickly picked it up and handed it back to her. “I liked him,” she said.



“Then he moved away,” she said.


“That’s all.”

“Lola, have you ever had sex?,” I heard myself saying.

“No,” she said.

She wasn’t accustomed to being asked personal questions. “Katulong lang ako,” she’d say. I’m only a servant. She often gave one- or two-word answers, and teasing out even the simplest story was a game of 20 questions that could last days or weeks.

Some of what I learned: She was mad at Mom for being so cruel all those years, but she nevertheless missed her. Sometimes, when Lola was young, she’d felt so lonely that all she could do was cry. I knew there were years when she’d dreamed of being with a man. I saw it in the way she wrapped herself around one large pillow at night. But what she told me in her old age was that living with Mom’s husbands made her think being alone wasn’t so bad. She didn’t miss those two at all. Maybe her life would have been better if she’d stayed in Mayantoc, gotten married, and had a family like her siblings. But maybe it would have been worse. Two younger sisters, Francisca and Zepriana, got sick and died. A brother, Claudio, was killed. What’s the point of wondering about it now? she asked. Bahala na was her guiding principle. Come what may. What came her way was another kind of family. In that family, she had eight children: Mom, my four siblings and me, and now my two daughters. The eight of us, she said, made her life worth living.

None of us was prepared for her to die so suddenly.

Her heart attack started in the kitchen while she was making dinner and I was running an errand. When I returned she was in the middle of it. A couple of hours later at the hospital, before I could grasp what was happening, she was gone—10:56 p.m. All the kids and grandkids noted, but were unsure how to take, that she died on November 7, the same day as Mom. Twelve years apart.

Lola made it to 86. I can still see her on the gurney. I remember looking at the medics standing above this brown woman no bigger than a child and thinking that they had no idea of the life she had lived. She’d had none of the self-serving ambition that drives most of us, and her willingness to give up everything for the people around her won her our love and utter loyalty. She’s become a hallowed figure in my extended family.

Going through her boxes in the attic took me months. I found recipes she had cut out of magazines in the 1970s for when she would someday learn to read. Photo albums with pictures of my mom. Awards my siblings and I had won from grade school on, most of which we had thrown away and she had “saved.” I almost lost it one night when at the bottom of a box I found a stack of yellowed newspaper articles I’d written and long ago forgotten about. She couldn’t read back then, but she’d kept them anyway.

Doods’s truck pulled up to a small concrete house in the middle of a cluster of homes mostly made of bamboo and plank wood. Surrounding the pod of houses: rice fields, green and seemingly endless. Before I even got out of the truck, people started coming outside.

Doods reclined his seat to take a nap. I hung my tote bag on my shoulder, took a breath, and opened the door.

“This way,” a soft voice said, and I was led up a short walkway to the concrete house. Following close behind was a line of about 20 people, young and old, but mostly old. Once we were all inside, they sat down on chairs and benches arranged along the walls, leaving the middle of the room empty except for me. I remained standing, waiting to meet my host. It was a small room, and dark. People glanced at me expectantly.

“Where is Lola?” A voice from another room. The next moment, a middle-aged woman in a housedress sauntered in with a smile. Ebia, Lola’s niece. This was her house. She gave me a hug and said again, “Where is Lola?”

I slid the tote bag from my shoulder and handed it to her. She looked into my face, still smiling, gently grasped the bag, and walked over to a wooden bench and sat down. She reached inside and pulled out the box and looked at every side. “Where is Lola?” she said softly. People in these parts don’t often get their loved ones cremated. I don’t think she knew what to expect. She set the box on her lap and bent over so her forehead rested on top of it, and at first I thought she was laughing (out of joy) but I quickly realized she was crying. Her shoulders began to heave, and then she was wailing—a deep, mournful, animal howl, like I once heard coming from Lola.

I hadn’t come sooner to deliver Lola’s ashes in part because I wasn’t sure anyone here cared that much about her. I hadn’t expected this kind of grief. Before I could comfort Ebia, a woman walked in from the kitchen and wrapped her arms around her, and then she began wailing. The next thing I knew, the room erupted with sound. The old people—one of them blind, several with no teeth—were all crying and not holding anything back. It lasted about 10 minutes. I was so fascinated that I barely noticed the tears running down my own face. The sobs died down, and then it was quiet again.

Ebia sniffled and said it was time to eat. Everybody started filing into the kitchen, puffy-eyed but suddenly lighter and ready to tell stories. I glanced at the empty tote bag on the bench, and knew it was right to bring Lola back to the place where she’d been born.

Tim Dettmers

Tim Dettmers

I am an informatics master student at the University of Lugano, Switzerland, currently during a research internship with the UCL Machine Reading Group where I am advised by Sebastian Riedel. In my work I focus on natural language understanding and more specifically, I work on deep learning for question answering and automatic knowledge base construction from raw text data. Before that I build my own GPU cluster and developed algorithms to speed up deep learning on GPU clusters. During my internship at Microsoft Research I worked on algorithms which make deep learning more memory efficient so that larger networks fit into GPU memory.

I also took part in Kaggle competitions where I have reached world rank 63, but currently research is more important to me than application of machine learning and deep learning.

In the past I studied applied mathematics at the Open University, and did a dual apprenticeship as Mathematical and Technical Software Developer where I worked in automation industry.

Besides deep learning, I am also very interested in understanding the human brain, human nature, the human condition and their evolution. In my spare time I like to study and think about fields aligned to these topics.

Garvity Payments

Garvity Payments

In 2015, Dan made headlines around the world when he announced to the entire Gravity team that he planned to raise the minimum wage for everyone to $70,000. He called the move a “moral imperative” to do the best you can for those you’re leading.

The response that followed his announcement was overwhelming. Dan’s move spurred hundreds of thousands to voice their support and inspired business owners across the country to employ similar pay increases for their teams.

For more facts and stories on the $70K minimum wage decision, go to thegravityof70k.com.



Recently, the New York Times ran a front-page story about the conditions for white-collar workers at Amazon. It revealed a workplace where abrupt firings are common, grown men and women cry at their desks, and people are scolded for not responding to e-mails after midnight. The story made clear how much things have changed in the American workforce. Once upon a time, it was taken for granted that the wealthier classes enjoyed a life of leisure on the backs of the proletariat. Today it is people in skilled trades who can most find reasonable hours coupled with good pay; the American professional is among those subject to humiliation and driven like a beast of burden.

No one thought things would be this way. The economist John Maynard Keynes famously forecast a three-hour workday, and in 1964 Life magazine devoted a two-part series to what it considered a “real threat” facing American society: the coming epidemic of too much leisure time. In “The Emptiness of Too Much Leisure,” it asserted that “some of the middle-of-the-road prophets of what automation is doing to our economy think that we are on the verge of a 30-hour week.” The follow-up was titled “The Task Ahead: How to Take Life Easy.”

Fifty years later, it’s fair to say that the looming leisure crisis has been licked. The work week at places like law firms, banks, and high-tech companies has steadily increased, to levels considered intolerable by many people. Indeed, in 2006, the top twenty per cent of earners were twice as likely to work more than fifty hours a week than the bottom twenty per cent, a reversal of historic conditions.

Just why this has happened is both a mystery and a paradox. The past fifty years have seen massive gains in productivity, the invention of countless labor-saving devices, and the mass entry of women into the formal workforce. If we assume that there is, to a certain degree, a fixed amount of work necessary for society to function, how can we at once be more productive, have more workers, and yet still be working more hours? Something else must be going on.

The question has proved a source of fascination for economists and writers, such as Brigid Schulte, a Washington Post reporter, who wrote a personal investigation of the question. (She ended up, in large part, blaming her husband, who wasn’t sharing equally in the burden of running their home.) As Elizabeth Kolbert has written, everyone agrees that there is no one simple answer to the question. Some people think that Americans just prefer work to leisure; a strong work ethic, according to this theory, has become a badge of honor for anyone with a college degree. If you’re busy, you seem important. There is also the pride that people can have in their work; they also find love and free food at workplaces, and go to conferences as a form of vacation. Others think the rise in work must somehow be related to inequality: as people at the top of the income ladder earn more money, each hour they work becomes more valuable. And there’s the theory that our needs and desires grow as we consume more, producing an even greater need to work.

What all of these explanations have in common is the idea that the answer comes from examining workers’ decisions and incentives. There’s something missing: the question of whether the American system, by its nature, resists the possibility of too much leisure, even if that’s what people actually want, and even if they have the means to achieve it. In other words, the long hours may be neither the product of what we really want nor the oppression of workers by the ruling class, the old Marxist theory. They may be the byproduct of systems and institutions that have taken on lives of their own and serve no one’s interests. That can happen if some industries have simply become giant make-work projects that trap everyone within them.

What counts as work, in the skilled trades, has some intrinsic limits; once a house or bridge is built, that’s the end of it. But in white-collar jobs, the amount of work can expand infinitely through the generation of false necessities—that is, reasons for driving people as hard as possible that have nothing to do with real social or economic needs. Consider the litigation system, in which the hours worked by lawyers at large law firms are a common complaint. If dispute resolution is the social function of the law, what we have is far from the most efficient way to reach fair or reasonable resolutions. Instead, modern litigation can be understood as a massive, socially unnecessary arms race, wherein lawyers subject each other to torturous amounts of labor just because they can. In older times, the limits of technology and a kind of professionalism created a natural limit to such arms races, but today neither side can stand down, lest it put itself at a competitive disadvantage.

A typical analysis blames greedy partners for crazy hours, but the irony is that the people at the top are often as unhappy and overworked as those at the bottom: it is a system that serves almost no one. Moreover, our many improvements in the technologies of productivity make the arms-race problem worse. The fact that employees are now always reachable eliminates what was once a natural barrier of sorts, the idea that work was something that happened during office hours or at the physical office. With no limits, work becomes like a football game where the whistle is never blown.

Litigation may be an extreme example, but I do not doubt that many other industries have their own arms races that create work that is of dubious necessity. The antidote is simple to prescribe but hard to achieve: it is a return to the goal of efficiency in work—fulfilling whatever needs we have, as a society, with the minimal effort required, while leaving the option of more work as a hobby for those who happen to love it. In this respect, it seems like no little irony that Amazon should be a brutal workplace when its ostensible guiding principle is making people’s lives better. There must be a better way.

portquiz.net: Outgoing port tester

portquiz.net: Outgoing port tester

Outgoing port tester

This server listens on all TCP ports, allowing you to test any outbound TCP port.
You have reached this page on port 80.
Your network allows you to use this port. (Assuming that your network is not doing advanced traffic filtering.)
Network service: http
Your outgoing IP:

Test a port using a command

$ telnet portquiz.net 80
Trying …
Connected to portquiz.net.
Escape character is ‘^]’.
$ nc -v portquiz.net 80
Connection to portquiz.net 80 port [tcp/daytime] succeeded!
$ curl portquiz.net:80
Port 80 test successful!
Your IP:
$ wget -qO- portquiz.net:80
Port 80 test successful!
Your IP:
# For Windows PowerShell users
PS C:\> Test-NetConnection -InformationLevel detailed -ComputerName portquiz.net -Port 80
Test a port using your browser

In your browser address bar: http://portquiz.net:XXXX

I got complains that portquiz is not working on port 445. My hosting company OVH is probably blocking this port. Sorry about that. Feel free to contact them. See my blog post and OVH forum post (french).

Your browser can block network ports normally used for purposes other than Web browsing. In this case you should use the telnet or netcat commands to test the port.

Please also note that this server uses some port for real services (22, 25), so testing with your browser on those ports will not work.



See also:
Blog post on this topic and How it works
Firebind, a commercial tester. javascript test
outPorts, a tiny program to test a range of ports using portquiz

Convolutional Neural Networks (CNNs): An Illustrated Explanation

Convolutional Neural Networks (CNNs): An Illustrated Explanation

Artificial Neural Networks (ANNs) are used everyday for tackling a broad spectrum of prediction and classification problems, and for scaling up applications which would otherwise require intractable amounts of data. ML has been witnessing a “Neural Revolution”1 since the mid 2000s, as ANNs found application in tools and technologies such as search engines, automatic translation, or video classification. Though structurally diverse, Convolutional Neural Networks (CNNs) stand out for their ubiquity of use, expanding the ANN domain of applicability from feature vectors to variable-length inputs.

The aim of this article is to give a detailed description of the inner workings of CNNs, and an account of the their recent merits and trends.

Table of Contents:

CNN Concepts
Input/Output Volumes
Filters (Convolution Kernels)
Kernel Operations Detailed
Receptive Field
The CNN Architecture
Convolutional Layer
The ReLu (Rectified Linear Unit) Layer
The Fully Connected Layer
CNN Design Principles
1The Neural Revolution is a reference to the period beginning 1982, when academic interest in the field of Neural Networks was invigorated by CalTech professor John J. Hopfield, who authored a research paper[1] that detailed the neural network architecture named after himself. The crucial breakthrough, however, occurred in 1986, when the backpropagation algorithm was proposed as such by David Rumelhart, Geoffrey E. Hinton and R.J. Williams [2]. For a history of neural networks, please see Andrey Kurenkov’s blog [3].


I would like to thank Adrian Scoica and Pedro Lopez for their immense patience and help with writing this piece. The sincerity of efforts and guidance that they’ve provided is ineffable. I’m forever inspired.


The modern Convolutional Neural Networks owe their inception to a well-known 1998 research paper[4] by Yann LeCun and Léon Bottou. In this highly instructional and detailed paper, the authors propose a neural architecture called LeNet 5 used for recognizing hand-written digits and words that established a new state of the art2 classification accuracy of 99.2% on the MNIST dataset[5].

According to the author’s accounts, CNNs are biologically-inspired models. The research investigations carried out by D. H. Hubel and T. N. Wiesel in their paper[6] proposed an explanation for the way in which mammals visually perceive the world around them using a layered architecture of neurons in the brain, and this in turn inspired engineers to attempt to develop similar pattern recognition mechanisms in computer vision.
The most popular application for CNNs in the recent times has been Image Analysis, but many researchers have also found other interesting and exciting ways to use them: from winning Go matches against human players([7], a related video [8]) to an innovative application in discovering new drugs by training over large quantities of molecular structure data of organic compounds[9].


A first question to answer with CNNs is why are they called Convolutional in the first place.

Convolution is a mathematical concept used heavily in Digital Signal Processing when dealing with signals that take the form of a time series. In lay terms, convolution is a mechanism to combine or “blend”[10] two functions of time3 in a coherent manner. It can be mathematically described as follows:

For a discrete domain of one variable:

CodeCogsEqn (1)

For a discrete domain of two variables:

CodeCogsEqn (2)

2A point to note here is the improvement is, in fact, modest. Classification accuracies greater than or equal to 99% on MNIST have been achieved using non-neural methods as well, such as K-Nearest Neighbours (KNN) or Support Vector Machines (SVM). For a list of ML methods applied and the respective classification accuracies attained, please refer to this[11] table.

3Or, for that matter, of another parameter.

Eq. 2 is perhaps more descriptive of what convolution truly is: a summation of pointwise products of function values, subject to traversal.

Though conventionally called as such, the operation performed on image inputs with CNNs is not strictly convolution, but rather a slightly modified variant called cross-correlation[10], in which one of the inputs is time-reversed:

2016-06-29 22_48_55-CNN_Blogpost_ – Google Docs

CNN Concepts

CNNs have an associated terminology and a set of concepts that is unique to them, and that sets them apart from other types of neural network architectures. The main ones are explained as follows:

Input/Output Volumes
CNNs are usually applied to image data. Every image is a matrix of pixel values. The range of values that can be encoded in each pixel depends upon its bit size. Most commonly, we have 8 bit or 1 Byte-sized pixels. Thus the possible range of values a single pixel can represent is [0, 255]. However, with coloured images, particularly RGB (Red, Green, Blue)-based images, the presence of separate colour channels (3 in the case of RGB images) introduces an additional ‘depth’ field to the data, making the input 3-dimensional. Hence, for a given RGB image of size, say 255×255 (Width x Height) pixels, we’ll have 3 matrices associated with each image, one for each of the colour channels. Thus the image in it’s entirety, constitutes a 3-dimensional structure called the Input Volume (255x255x3).

Figure 1: The cross-section of an input volume of size: 4 x 4 x 3. It comprises of the 3 Colour channel matrices of the input image.


Just as its literal meaning implies, a feature is a distinct and useful observation or pattern obtained from the input data that aids in performing the desired image analysis. The CNN learns the features from the input images. Typically, they emerge repeatedly from the data to gain prominence. As an example, when performing Face Detection, the fact that every human face has a pair of eyes will be treated as a feature by the system, that will be detected and learned by the distinct layers. In generic object classification, the edge contours of the objects serve as the features.

Filters (Convolution Kernels)

A filter (or kernel) is an integral component of the layered architecture.

Generally, it refers to an operator applied to the entirety of the image such that it transforms the information encoded in the pixels. In practice, however, a kernel is a smaller-sized matrix in comparison to the input dimensions of the image, that consists of real valued entries.

The kernels are then convolved with the input volume to obtain so-called ‘activation maps’. Activation maps indicate ‘activated’ regions, i.e. regions where features specific to the kernel have been detected in the input. The real values of the kernel matrix change with each learning iteration over the training set, indicating that the network is learning to identify which regions are of significance for extracting features from the data.

Kernel Operations Detailed

The exact procedure for convolving a Kernel (say, of size 16 x 16) with the input volume (a 256 x 256 x 3 sized RGB image in our case) involves taking patches from the input image of size equal to that of the kernel (16 x 16), and convolving (or calculating the dot product) between the values in the patch and those in the kernel matrix.

The convolved value obtained by summing the resultant terms from the dot product forms a single entry in the activation matrix. The patch selection is then slided (towards the right, or downwards when the boundary of the matrix is reached) by a certain amount called the ‘stride’ value, and the process is repeated till the entire input image has been processed. The process is carried out for all colour channels. For normalization purposes, we divide the calculated value of the activation matrix by the sum of values in the kernel matrix.

The process is demonstrated in Figure 2, using a toy example consisting of a 3-channel 4×4-pixels input image and a 3×3 kernel matrix. Note that:

pixels are numbered from 1 in the example;
the values in the activation map are normalized to ensure the same intensity range between the input volume and the output volume. Hence, for normalization, we divide the calculated value for the ‘red’ channel by 2 (the sum of values in the kernel matrix);
we assume the same kernel matrix for all the three channels, but it is possible to have a separate kernel matrix for each colour channel;
for a more detailed and intuitive explanation of the convolution operation, you can refer to the excellent blog-posts by Chris Olah[12] and by Tim Dettmers[13].

Figure 2: The convolution value is calculated by taking the dot product of the corresponding values in the Kernel and the channel matrices. The current path is indicated by the red-coloured, bold outline in the Input Image volume. Here, the entry in the activation matrix is calculated as:

CodeCogsEqn (6)

Receptive Field

It is impractical to connect all neurons with all possible regions of the input volume. It would lead to too many weights to train, and produce too high a computational complexity. Thus, instead of connecting each neuron to all possible pixels, we specify a 2 dimensional region called the ‘receptive field[14]’ (say of size 5×5 units) extending to the entire depth of the input (5x5x3 for a 3 colour channel input), within which the encompassed pixels are fully connected to the neural network’s input layer. It’s over these small regions that the network layer cross-sections (each consisting of several neurons (called ‘depth columns’)) operate and produce the activation map.


Zero-padding refers to the process of symmetrically adding zeroes to the input matrix. It’s a commonly used modification that allows the size of the input to be adjusted to our requirement. It is mostly used in designing the CNN layers when the dimensions of the input volume need to be preserved in the output volume.


Figure 3: A zero-padded 4 x 4 matrix becomes a 6 x 6 matrix.


In CNNs, the properties pertaining to the structure of layers and neurons, such spatial arrangement and receptive field values, are called hyperparameters. Hyperparameters uniquely specify layers. The main CNN hyperparameters are receptive field (R), zero-padding (P), the input volume dimensions (Width x Height x Depth, or W x H x D ) and stride length (S).

The CNN Architecture

Now that we are familiar with the CNN terminology, let’s go on ahead and study the CNN architecture in detail.

The architecture of a typical CNN is composed of multiple layers where each layer performs a specific function of transforming its input into a useful representation. There are 3 major types of layers that are commonly observed in complex neural network architectures:

Convolutional Layer
Also referred to as Conv. layer, it forms the basis of the CNN and performs the core operations of training and consequently firing the neurons of the network. It performs the convolution operation over the input volume as specified in the previous section, and consists of a 3-dimensional arrangement of neurons (a stack of 2-dimensional layers of neurons, one for each channel depth).


Figure 4: A 3-D representation of the Convolutional layer with 3 x 3 x 4 = 36 neurons.

Each neuron is connected to a certain region of the input volume called the receptive field (explained in the previous section). For example, for an input image of dimensions 28x28x3, if the receptive field is 5 x 5, then each neuron in the Conv. layer is connected to a region of 5x5x3 (the region always comprises the entire depth of the input, i.e. all the channel matrices) in the input volume. Hence each neuron will have 75 weighted inputs. For a particular value of R (receptive field), we have a cross-section of neurons entirely dedicated to taking inputs from this region. Such a cross-section is called a ‘depth column’. It extends to the entire depth of the Conv. layer.

For optimized Conv. layer implementations, we may use a Shared Weights model that reduces the number of unique weights to train and consequently the matrix calculations to be performed per layer. In this model, each ‘depth slice’ or a single 2-dimensional layer of neurons in the Conv architecture all share the same weights. The caveat with parameter sharing is that it doesn’t work well with images that encompass a spatially centered structure (such as face images), and in applications where we want the distinct features of the image to be detected in spatially different locations of the layer.


Figure 5: Concept of Receptive Field.

We must keep in mind though that the network operates in the same way that a feed-forward network would: the weights in the Conv layers are trained and updated in each learning iteration using a Back-propagation algorithm extended to be applicable to 3-dimensional arrangements of neurons.

The ReLu (Rectified Linear Unit) Layer

ReLu refers to the Rectifier Unit, the most commonly deployed activation function for the outputs of the CNN neurons. Mathematically, it’s described as:

CodeCogsEqn (3)

Unfortunately, the ReLu function is not differentiable at the origin, which makes it hard to use with backpropagation training. Instead, a smoothed version called the Softplus function is used in practice:

CodeCogsEqn (4)

The derivative of the softplus function is the sigmoid function, as mentioned in a prior blog post.

CodeCogsEqn (5)

The Pooling Layer

The pooling layer is usually placed after the Convolutional layer. Its primary utility lies in reducing the spatial dimensions (Width x Height) of the Input Volume for the next Convolutional Layer. It does not affect the depth dimension of the Volume.

The operation performed by this layer is also called ‘down-sampling’, as the reduction of size leads to loss of information as well. However, such a loss is beneficial for the network for two reasons:

the decrease in size leads to less computational overhead for the upcoming layers of the network;
it work against over-fitting.
Much like the convolution operation performed above, the pooling layer takes a sliding window or a certain region that is moved in stride across the input transforming the values into representative values. The transformation is either performed by taking the maximum value from the values observable in the window (called ‘max pooling’), or by taking the average of the values. Max pooling has been favoured over others due to its better performance characteristics.

The operation is performed for each depth slice. For example, if the input is a volume of size 4x4x3, and the sliding window is of size 2×2, then for each color channel, the values will be down-sampled to their representative maximum value if we perform the max pooling operation.

No new parameters are introduced in the matrix by this operation. The operation can be thought of as applying a function over input values, taking fixed sized portions at a time, with the size, modifiable as a parameter. Pooling is optional in CNNs, and many architectures do not perform pooling operations.


Figure 6: The Max-Pooling operation can be observed in sub-figures (i), (ii) and (iii) that max-pools the 3 colour channels for an example input volume for the pooling layer. The operation uses a stride value of [2, 2]. The dark and red boundary regions describe the window movement. Sub-figure (iv) shows the operation applied for a stride value of [1,1], resulting in a 3×3 matrix Here we observe overlap between regions.

The Fully Connected Layer

The Fully Connected layer is configured exactly the way its name implies: it is fully connected with the output of the previous layer. Fully-connected layers are typically used in the last stages of the CNN to connect to the output layer and construct the desired number of outputs.

CNN Design Principles

Given the aforementioned building blocks, the last detail before implementing a CNN is to specify its design end to end, and to decide on the layer dimensions of the Convolutional layers.

A quick and dirty empirical formula[15] for calculating the spatial dimensions of the Convolutional Layer as a function of the input volume size and the hyperparameters we discussed before can be written as follows:

For each (ith) dimension of the input volume, pick:


where is the (ith) input dimension, R is the receptive field value, P is the padding value, and S is the value of the stride. Note that the formula does not rely on the depth of the input.

To better understand better how it works, let’s consider the following example:

Let the dimensions of the input volume be 288x288x3, the stride value be 2 (both along horizontal and vertical directions).
Now, since WIn=288 and S = 2, (2.P – R) must be an even integer for the calculated value to be an integer. If we set the padding to 0 and R = 4, we get WOut=(288-4+2.0)/2+1 =284/2 + 1 = 143. As the spatial dimensions are symmetrical (same value for width and height), the output dimensions are going to be: 143 x 143 x K, where K is the depth of the layer. K can be set to any value, with increasing values for every Conv. layer added. For larger networks values of 512 are common.
The output volume from a Conv. layer either has the same dimensions as that of the Conv. layer (143x143x2 for the example considered above), or the same as that of the input volume (288x288x3 for the example above).
The generic arrangement of layers can thus be summarized as follows[15]:


Where N usually takes values between 0 and 3, M >= 0 and K∈[0,3).

The expression indicates multiple layers, with or without per layer-Pooling. The final layer is the fully-connected output layer. See [8] for more case-studies of CNN architectures, as well as a detailed discussion of layers and hyper-parameters.


CNNs showcase the awesome levels of control over performance that can be achieved by making effective use of theoretical and mathematical insights. Many real world problems are being efficiently tackled using CNNs, and MNIST represents a simple, “Hello World”-type use-case of this technique. More complex problems such as object and image recognition require the use of deep neural networks with millions of parameters to obtain state-of-the-art results. CIFAR-10 is a good problem to solve in this domain, and it was first solved by Alex Krizhevsky et al.[16] in 2009. You can read through the technical report and try and grasp the approach before making way to the TensorFlow tutorial that solves the same problem[17].

Furthermore, applications are not limited to computer vision. The most recent win of Google’s AlphaGo Project over Lee Sedol in the Go game series relied on a CNN at its core. The self-driving cars which, in the coming years, will arguably become a regular sight on our streets, rely on CNNs for steering[18]. Google even held an art-show[19] for imagery created by its DeepDream project that showcased beautiful works of art created by visualizing the transformations of the network!

Thus a Machine Learning researcher or engineer in today’s world can rejoice at the technological melange of techniques at her disposal, among which an in-depth understanding of CNNs is both indispensable and empowering.


[1] Hopfield, John J. “Neural networks and physical systems with emergent collective computational abilities.” Proceedings of the national academy of sciences 79.8 (1982): 2554-2558.[http://www.pnas.org/content/79/8/2554.abstract]

[2] Rumelhart, D. E., Hinton, G. E., and Williams, R. J. (1986). Learning representations by back-propagating errors. Nature, 323, 533–536.[http://www.iro.umontreal.ca/~vincentp/ift3395/lectures/backprop_old.pdf]

[3] Andrey Kurenkov, “A brief History of Neural Nets and Deep Learning”.[http://www.andreykurenkov.com/writing/a-brief-history-of-neural-nets-and-deep-learning/]

[4] LeCun, Yann, et al. “Gradient-based learning applied to document recognition.” Proceedings of the IEEE 86.11 (1998): 2278-2324.

[5] The MNIST database of handwritten digits

[6] Hubel, David H., and Torsten N. Wiesel. “Receptive fields and functional architecture of monkey striate cortex.” The Journal of physiology 195.1 (1968): 215-243.

[7] Alpha Go video by Nature. [http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234]

[8] Clark, Christopher, and Amos Storkey. “Teaching deep convolutional neural networks to play go.” arXiv preprint arXiv:1412.3409 (2014).[http://arxiv.org/pdf/1412.3409.pdf]

[9] Wallach, Izhar, Michael Dzamba, and Abraham Heifets. “AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery.” arXiv preprint arXiv:1510.02855 (2015).


[10] Weisstein, Eric W. “Convolution.” From MathWorld — A Wolfram Web Resource. [http://mathworld.wolfram.com/Convolution.html]

[11] Table of classification accuracies attained over MNIST. [https://en.wikipedia.org/wiki/MNIST_database#Performance]

[12] Chris Olah, “Understanding Convolutions”. [http://colah.github.io/posts/2014-07-Understanding-Convolutions/]

[13] Tim Dettmers, “Understanding Convolution In Deep Learning”.[http://timdettmers.com/2015/03/26/convolution-deep-learning/]

[14] TensorFlow Documentation: Convolution [https://www.tensorflow.org/versions/r0.7/api_docs/python/nn.html#convolution]

[15] Andrej Karpathy, “CS231n: Convolutional Neural Networks for Visual Recognition” [http://cs231n.github.io/convolutional-networks/]

[16] Krizhevsky, Alex, and Geoffrey Hinton. “Learning multiple layers of features from tiny images.” (2009).[http://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf]

[17] TensorFlow: Convolutional Networks.[https://www.tensorflow.org/versions/r0.7/tutorials/deep_cnn/index.html#cifar-10-model]

[18] Google Deepmind’s AlphaGo: How it works. [https://www.tastehit.com/blog/google-deepmind-alphago-how-it-works/]

[19] An Empirical Evaluation of Deep Learning on Highway Driving.[http://arxiv.org/pdf/1504.01716.pdf]

[20] Inside Google’s First DeepDream Art Project. [http://www.fastcodesign.com/3057368/inside-googles-first-deepdream-art-show/11]

This entry was posted in machine learning, Uncategorized, xrds by Abhineet Saxena. Bookmark the permalink.