The CRA-W DLS program sponsored workshop on deep learning was held at Pace University, Seidenberg School on Friday April 28th. The key speaker Anna Goldie, is a software engineer in Google Brain Team. Her research focuses on language understanding, question-answering, and conversational modeling. In the morning section, Anna gave a presentation on the basics of deep learning theory and algorithms. In the afternoon section, Anna invited her NY Google colleague Josh Gordon as a helper and provided a hands-on workshop with exercises and real data sets to illustrate how to use Google TensorFlow to train one's own deep neural network. In the afternoon section, attendees were able to practice using TensorFlow from the simplest example to more complicated date sets. Anna and Josh came to attendees' seats to help people in a one-to-one manner.

The afternoon workshop was followed by graduate school panel discussion. Since the attendees were mainly graduate students, the topics included both graduate school application and PhD program application, as well as Women in Tech Club at Pace University, and a presentation about CRA-W workshop for female students shared by Vanessa Rene who had recently attended the CRA-W workshop at DC.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

SOMA Robotics Security Robot Prototype 2. This one scared people even more than prototype 1, When they write "it reminds me of a velociraptor" on the follow up survey, I think we can safely rule out this form factor even if I thought it was cool, especially hung upside down from the ceiling.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Bangle.js is the first hackable open source JS and TensorFlow-driven smartwatch. NodeConf EU 2019. Kilkenny, Ireland. November 2019. Photo by Nico Kaiser

This installation creates a situation where the fate of a colony of living houseflies is determined by the accuracy of artificial-intelligence software. The installation uses the TensorFlow machine-learning image-recognition library to classify images of live houseflies. As the flies fly and land in front of a camera, their image is captured. The captured image is classified by the image-recognition software and a list of guessed items is ranked one through five. Each of the items is assigned a percentage based on how likely the software thinks the listed item is what it sees. If “fly” is ranked number one on the list, a pump delivers water and nutrients to the colony based on the percentage of the ranking. If “fly” is not ranked number one the pump does not deliver water and nutrients to the colony. The system is set up to run indefinitely with an indeterminate outcome.

credit: David Bowen

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Bangle.js is the first hackable open source JS and TensorFlow-driven smartwatch. NodeConf EU 2019. Kilkenny, Ireland. November 2019. Photo by Nico Kaiser

Bangle.js is the first hackable open source JS and TensorFlow-driven smartwatch. NodeConf EU 2019. Kilkenny, Ireland. November 2019. Photo by Nico Kaiser

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

This AI and Deep learning course offers practical and task-oriented training using TensorFlow and Keras on Python platform. This is a specialization course which will help you to get a break into AI and Deep Learning domain. Enroll today. For more visit here www.analytixlabs.co.in/ai-deep-learning-training-with-python

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Jeff Dean sat for this portrait in Google’s Building 43 on an August afternoon in 2025. The room was quiet, the light falling in a way that made his eyes appear both steady and amused. He has the look of someone who has been through countless problem-solving sessions but still finds joy in the process.

Dean’s story begins in Minnesota, where he studied computer science before heading west for graduate school at the University of Washington. There he focused on compilers, writing work that was highly technical but deeply practical. It was less about abstract beauty and more about making computers run faster, squeezing efficiency out of systems that always seemed too limited. That mindset would shape the rest of his career.

His first real laboratory was Digital Equipment Corporation’s Western Research Lab in Palo Alto. At DEC he was surrounded by engineers who cared about elegance but never at the expense of reliability. Those years gave him a lasting respect for craftsmanship in software. He has spoken of how formative it was to be in a place where even the smallest decisions about performance and structure mattered.

When he arrived at Google in 1999, the company had fewer than a hundred employees. Search was already straining the capacity of existing systems. Dean and his longtime collaborator Sanjay Ghemawat began sketching solutions on whiteboards, filling them with boxes and arrows that hinted at ways to divide work across thousands of machines. Out of those sessions came MapReduce, which allowed Google to process massive data sets in a fraction of the time. Bigtable followed, giving the company a storage system that could keep pace with its ambitions. Engineers still remember the first time they saw queries finish in hours instead of days.

Stories about Dean inside Google often return to the way he approaches code reviews. He is known to mark up a colleague’s code with detailed comments, sometimes line by line, always pushing for clarity. Yet the tone is never dismissive. Younger engineers recall feeling surprised that someone of his stature took their work so seriously. It gave them confidence to tackle harder problems.

By the middle of the 2010s Dean turned his focus to artificial intelligence. As head of Google Brain, he encouraged the team to take bold steps, even when success was uncertain. TensorFlow, the open-source library they released, was built with that same spirit. It made sophisticated machine learning accessible, and within a few years it was being used by researchers in medicine, climate science, and language technology. Dean liked to point out that the best ideas often came from unexpected places once the tools were in wide circulation.

His colleagues describe a leadership style that is more invitation than command. In meetings he often starts with a quiet question that reframes the problem, steering the discussion toward fundamentals. He is patient with complexity, willing to sit through hours of debate if it means arriving at a solution that will endure. His amusement shows itself in small ways, often in a half-smile when a tough question lands on the table.

Outside recognition has come steadily. Election to the National Academy of Engineering confirmed his standing as one of the country’s most influential technologists. Awards have followed, but inside Google what people talk about most is his constancy. He shows up, he listens, and he remains deeply engaged with the work itself.

To photograph him is to see both sides at once. His arms are folded, his expression serious, yet his eyes suggest he is ready to lean forward into the next idea. The systems he has helped build are vast and nearly invisible, woven into daily life for billions of people. But in person he is disarmingly affable, a reminder that behind the abstractions of code and scale is a human being who still loves to puzzle out how things work.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

The CRA-W DLS program sponsored workshop on deep learning was held at Pace University, Seidenberg School on Friday April 28th. The key speaker Anna Goldie, is a software engineer in Google Brain Team. Her research focuses on language understanding, question-answering, and conversational modeling. In the morning section, Anna gave a presentation on the basics of deep learning theory and algorithms. In the afternoon section, Anna invited her NY Google colleague Josh Gordon as a helper and provided a hands-on workshop with exercises and real data sets to illustrate how to use Google TensorFlow to train one's own deep neural network. In the afternoon section, attendees were able to practice using TensorFlow from the simplest example to more complicated date sets. Anna and Josh came to attendees' seats to help people in a one-to-one manner.

The afternoon workshop was followed by graduate school panel discussion. Since the attendees were mainly graduate students, the topics included both graduate school application and PhD program application, as well as Women in Tech Club at Pace University, and a presentation about CRA-W workshop for female students shared by Vanessa Rene who had recently attended the CRA-W workshop at DC.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Photos by Stan Olszewski. Editorial and blog use in articles about O'Reilly TensorFlow World, as well as personal

and non-commercial use, is permitted with attribution alongside the image and a link back to this photo page where possible.

Devoxx 2018 - From 0 to Deep Learning in 3 hours

Neural Networks, Deep Learning: there's not a week without a thunderous announcement claiming that Artificial Intelligence limits are once again pushed further.

Doesn't that make you wonder what's actually behind all that? Would you like to know how this all works and how we'll end up having all of our jobs stolen by soulless machines?

Sure, but you've heard that this implies tons of matrix calculus, algebra that doesn't always seem to be that linear, partial derivatives and, come on, if we chose programming rather than pursuing a PhD in Mathematics, we knew what we were doin'!

Okay, give me 3 hours and I will tell you, in very simple words, without any prior knowledge required, step by step baby, how neural networks work in real life, how we get them to learn very useful tricks such as how to differentiate (pun intended) cats from dogs or recognize traffic signs in real time, and how we use these networks in practice.

And we'll see that, thanks to libraries such as TensorFlow or Keras, a few lines of code are enough to do wonders...

Ready?

www.youtube.com/watch?v=gy6cLz4ra8E

( Devoxx 2018

Tous les slides sont proprietes de leurs auteurs.

All slides are properties of their authors. )

Bangle.js is the first hackable open source JS and TensorFlow-driven smartwatch. NodeConf EU 2019. Kilkenny, Ireland. November 2019. Photo by Nico Kaiser