Today we’re joined by Stefan Lee, assistant professor at the school of electrical engineering and computer science at Oregon State University. Stefan, who we sat down with at NeurIPS this past winter, is focused on the development of agents that can perceive their environment and communicate their understanding with humans in order to coordinate their actions to achieve mutual goals.  In our conversation, we focus on his paper <a href= "https://arxiv.org/abs/1908.02265">ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks</a>, a model for learning joint representations of image content and natural language. We talk through the development and training process for this model, the adaptation of the training process to incorporate additional visual information to BERT models, where this research leads from the perspective of integration between visual and language tasks and finally, we discuss the importance of visual grounding. Check out the complete show notes page at <a href= "https://twimlai.com/talk/358">twimlai.com/talk/358</a>.

Today we’re joined by Stefan Lee, assistant professor at the school of electrical engineering and computer science at Oregon State University. Stefan, who we sat down with at NeurIPS this past winter, is focused on the development of agents that can perceive their environment and communicate their understanding with humans in order to coordinate their actions to achieve mutual goals.  In our conversation, we focus on his paper ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks, a model for learning joint representations of image content and natural language. We talk through the development and training process for this model, the adaptation of the training process to incorporate additional visual information to BERT models, where this research leads from the perspective of integration between visual and language tasks and finally, we discuss the importance of visual grounding. Check out the complete show notes page at twimlai.com/talk/358.

This Week in Machine Learning & Artificial Intelligence (AI) Podcast

Learning Visiolinguistic Representations with ViLBERT w/ Stefan Lee - #358

Today we’re joined by Jürgen Schmidhuber, Co-Founder and Chief Scientist of NNAISENSE, the Scientific Director at IDSIA, as well as a Professor of AI at USI and SUPSI in Switzerland. Jürgen’s lab is well known for creating the Long Short-Term Memory (LSTM) network which has become a prevalent neural network, used commonly devices such as smartphones, which we discuss in detail in our first conversation with Jürgen back in 2017. In this conversation, we dive into some of Jürgen’s more recent work, including his recent paper, <a href= "https://arxiv.org/abs/1912.02875">Reinforcement Learning Upside Down: Don’t Predict Rewards — Just Map Them to Actions</a>. Check out the show notes page at <a href= "https://twimlai.com/talk/357">twimlai.com/talk/357</a>.

Today we’re joined by Jürgen Schmidhuber, Co-Founder and Chief Scientist of NNAISENSE, the Scientific Director at IDSIA, as well as a Professor of AI at USI and SUPSI in Switzerland. Jürgen’s lab is well known for creating the Long Short-Term Memory (LSTM) network which has become a prevalent neural network, used commonly devices such as smartphones, which we discuss in detail in our first conversation with Jürgen back in 2017. In this conversation, we dive into some of Jürgen’s more recent work, including his recent paper, Reinforcement Learning Upside Down: Don’t Predict Rewards — Just Map Them to Actions. Check out the show notes page at twimlai.com/talk/357.

Upside-Down Reinforcement Learning with Jürgen Schmidhuber - #357

<div class="c-message_kit__gutter"> <div class="c-message_kit__gutter__right" data-qa= "message_content"> <div class= "c-message_kit__blocks c-message_kit__blocks--rich_text"> <div class= "c-message__message_blocks c-message__message_blocks--rich_text"> <div class="p-block_kit_renderer" data-qa="block-kit-renderer"> <div class= "p-block_kit_renderer__block_wrapper p-block_kit_renderer__block_wrapper--first"> <div class="p-rich_text_block" dir="auto"> <div class="p-rich_text_section">Today we're joined by Beidi Chen, PhD student at Rice University. Beidi is part of the team that developed a cheaper, algorithmic, CPU alternative to state-of-the-art GPU machines. They presented their findings at NeurIPS 2019 and have since gained a lot of attention for their paper, SLIDE: In Defense of Smart Algorithms Over Hardware Acceleration for Large-Scale Deep Learning Systems. In this interview, Beidi shares how the team took a new look at deep learning with the case of extreme classification by turning it into a search problem and using locality-sensitive hashing.</div> </div> </div> </div> </div> </div> </div> </div> <div class="c-message_actions__container c-message__actions" role= "group" aria-label="Message actions" data-qa="message-actions">  </div> <div class="c-message_actions__container c-message__actions" role= "group" aria-label="Message actions" data-qa="message-actions"> Check out the complete show notes at twimlai.com/talk/356. </div>

 Today we're joined by Beidi Chen, PhD student at Rice University. Beidi is part of the team that developed a cheaper, algorithmic, CPU alternative to state-of-the-art GPU machines. They presented their findings at NeurIPS 2019 and have since gained a lot of attention for their paper, SLIDE: In Defense of Smart Algorithms Over Hardware Acceleration for Large-Scale Deep Learning Systems. In this interview, Beidi shares how the team took a new look at deep learning with the case of extreme classification by turning it into a search problem and using locality-sensitive hashing.   Check out the complete show notes at twimlai.com/talk/356. 

SLIDE: Smart Algorithms over Hardware Acceleration for Large-Scale Deep Learning with Beidi Chen - #356

Today we're joined by Sergey Levine, an Assistant Professor in the Department of Electrical Engineering and Computer Science at UC Berkeley. We last heard from Sergey back in 2017, where we explored Deep Robotic Learning. We caught up with Sergey at NeurIPS 2019, where Sergey and his team presented 12 different papers -- which means a lot of ground to cover! Sergey and his lab’s recent efforts have been focused on contributing to a future where machines can be “out there in the real world, learning continuously through their own experience.” Sergey shares how many of the papers presented at the most recent NeurIPS conference are working to make that happen. Some of the major developments have been in the research fields of model-free reinforcement learning, causality and imitation learning, and offline reinforcement learning. Check out the complete show notes page at <a href= "https://twimlai.com/talk/355">twimlai.com/talk/355</a>.

Today we're joined by Sergey Levine, an Assistant Professor in the Department of Electrical Engineering and Computer Science at UC Berkeley. We last heard from Sergey back in 2017, where we explored Deep Robotic Learning. We caught up with Sergey at NeurIPS 2019, where Sergey and his team presented 12 different papers -- which means a lot of ground to cover! Sergey and his lab’s recent efforts have been focused on contributing to a future where machines can be “out there in the real world, learning continuously through their own experience.” Sergey shares how many of the papers presented at the most recent NeurIPS conference are working to make that happen. Some of the major developments have been in the research fields of model-free reinforcement learning, causality and imitation learning, and offline reinforcement learning. Check out the complete show notes page at twimlai.com/talk/355.

Advancements in Machine Learning with Sergey Levine - #355

Imagine spending years learning ML from the ground up, from its theoretical foundations, but still feeling like you didn’t really know how to apply it. That’s where David Odaibo found himself in 2015, after the second year of his PhD. David’s solution was Kaggle, a popular platform for data science competitions. Fast forward four years, and David is now a Kaggle Grandmaster, the highest designation, with particular accomplishment in computer vision competitions. Having completed his degree last year, he is currently co-founder and CTO of Analytical AI, a company that grew out of one of his recent Kaggle successes. David has a background in deep learning and medical imaging–something he shares with his brother, Stephen Odaibo, who we interviewed last year about his work in <a href= "https://twimlai.com/twiml-talk-284-retinal-image-generation-for-disease-discovery-with-stephen-odaibo/"> Retinal Image Generation for Disease Discovery</a>. Check out the full article and interview at <a href= "https://twimlai.com/talk/354">twimlai.com/talk/354</a>

Imagine spending years learning ML from the ground up, from its theoretical foundations, but still feeling like you didn’t really know how to apply it. That’s where David Odaibo found himself in 2015, after the second year of his PhD. David’s solution was Kaggle, a popular platform for data science competitions. Fast forward four years, and David is now a Kaggle Grandmaster, the highest designation, with particular accomplishment in computer vision competitions. Having completed his degree last year, he is currently co-founder and CTO of Analytical AI, a company that grew out of one of his recent Kaggle successes. David has a background in deep learning and medical imaging–something he shares with his brother, Stephen Odaibo, who we interviewed last year about his work in Retinal Image Generation for Disease Discovery. Check out the full article and interview at twimlai.com/talk/354

Secrets of a Kaggle Grandmaster with David Odaibo - #354

Predicting the future of science, particularly physics, is the task that Matteo Chinazzi, an associate research scientist at Northeastern University focused on in his paper <a href= "https://epjdatascience.springeropen.com/track/pdf/10.1140/epjds/s13688-019-0210-z"> Mapping the Physics Research Space: a Machine Learning Approach</a>, along with co-authors including former TWIML AI Podcast guest <a href= "https://twimlai.com/twiml-talk-048-word2vec-friends-bruno-goncalves/"> Bruno Gonçalves</a>. In addition to predicting the trajectory of physics research, Matteo is also active in the computational epidemiology field. His work in that area involves building simulators that can model the spread of diseases like Zika or the seasonal flu at a global scale.  Check out our full article on this episode at twimlai.com/talk/353.

Predicting the future of science, particularly physics, is the task that Matteo Chinazzi, an associate research scientist at Northeastern University focused on in his paper Mapping the Physics Research Space: a Machine Learning Approach, along with co-authors including former TWIML AI Podcast guest Bruno Gonçalves. In addition to predicting the trajectory of physics research, Matteo is also active in the computational epidemiology field. His work in that area involves building simulators that can model the spread of diseases like Zika or the seasonal flu at a global scale.  Check out our full article on this episode at twimlai.com/talk/353.

NLP for Mapping Physics Research with Matteo Chinazzi - #353

The unfortunate reality is that many of the most commonly used machine learning metrics don't account for the complex trade-offs that come with real-world decision making. This is one of the challenges that today’s guest, Sanmi Koyejo has dedicated his research to address. Sanmi is an assistant professor at the Department of Computer Science at the University of Illinois, where he applies his background in cognitive science, probabilistic modeling, and Bayesian inference to pursue his research which focuses broadly on “adaptive and robust machine learning.” Check out the full episode write-up at <a href= "https://twimlai.com/talk/352">twimlai.com/talk/352</a>.

The unfortunate reality is that many of the most commonly used machine learning metrics don't account for the complex trade-offs that come with real-world decision making. This is one of the challenges that today’s guest, Sanmi Koyejo has dedicated his research to address. Sanmi is an assistant professor at the Department of Computer Science at the University of Illinois, where he applies his background in cognitive science, probabilistic modeling, and Bayesian inference to pursue his research which focuses broadly on “adaptive and robust machine learning.” Check out the full episode write-up at twimlai.com/talk/352.

Metric Elicitation and Robust Distributed Learning with Sanmi Koyejo - #352

Today we’re joined by Ilias Diakonikolas, faculty in the CS department at the University of Wisconsin-Madison, and author of the paper Distribution-Independent PAC Learning of Halfspaces with Massart Noise, which was the recipient of the NeurIPS 2019 Outstanding Paper award. The paper, which focuses on high-dimensional robust learning, is regarded as the first progress made around distribution-independent learning with noise since the 80s. In our conversation, we explore robustness in machine learning, problems with corrupt data in high-dimensional settings, and of course, a deep dive into the paper.  Check out our full write up on the paper and the interview at <a href= "https://twimlai.com/talk/351">twimlai.com/talk/351</a>.

Today we’re joined by Ilias Diakonikolas, faculty in the CS department at the University of Wisconsin-Madison, and author of the paper Distribution-Independent PAC Learning of Halfspaces with Massart Noise, which was the recipient of the NeurIPS 2019 Outstanding Paper award. The paper, which focuses on high-dimensional robust learning, is regarded as the first progress made around distribution-independent learning with noise since the 80s. In our conversation, we explore robustness in machine learning, problems with corrupt data in high-dimensional settings, and of course, a deep dive into the paper.  Check out our full write up on the paper and the interview at twimlai.com/talk/351.

High-Dimensional Robust Statistics with Ilias Diakonikolas - #351

Today we’re joined by Kamran Khan, founder & CEO of BlueDot, and professor of medicine and public health at the University of Toronto. BlueDot, a digital health company with a focus on surveilling global infectious disease outbreaks, has been the recipient of a lot of attention for being the first to publicly warn about the coronavirus that started in Wuhan. How did the company’s system of algorithms and data processing techniques help flag the potential dangers of the disease? In this interview, Kamran talks us through how the technology works, its limits, and the motivation behind the work.  Check out our new and improved show notes article at <a href="https://twimlai.com/talk/350" target="_blank" rel="noopener">twimlai.com/talk/350</a>.

Today we’re joined by Kamran Khan, founder & CEO of BlueDot, and professor of medicine and public health at the University of Toronto. BlueDot, a digital health company with a focus on surveilling global infectious disease outbreaks, has been the recipient of a lot of attention for being the first to publicly warn about the coronavirus that started in Wuhan. How did the company’s system of algorithms and data processing techniques help flag the potential dangers of the disease? In this interview, Kamran talks us through how the technology works, its limits, and the motivation behind the work.  Check out our new and improved show notes article at twimlai.com/talk/350.

How AI Predicted the Coronavirus Outbreak with Kamran Khan - #350

Today we’re joined by Emmanuel Ameisen, machine learning engineer at Stripe, and author of the recently published book “Building Machine Learning Powered Applications; Going from Idea to Product.” In our conversation, we discuss structuring end-to-end machine learning projects, debugging and explainability in the context of models, the various types of models covered in the book, and the importance of post-deployment monitoring.  Check out our full show notes article at <a href= "https://twimlai.com/talk/349">twimlai.com/talk/349</a>.

Today we’re joined by Emmanuel Ameisen, machine learning engineer at Stripe, and author of the recently published book “Building Machine Learning Powered Applications; Going from Idea to Product.” In our conversation, we discuss structuring end-to-end machine learning projects, debugging and explainability in the context of models, the various types of models covered in the book, and the importance of post-deployment monitoring.  Check out our full show notes article at twimlai.com/talk/349.

Turning Ideas into ML Powered Products with Emmanuel Ameisen - #349

Today we’re joined by Abeba Birhane, PhD Student at University College Dublin and author of the recent paper <a href= "https://arxiv.org/pdf/1912.07376v1.pdf">Algorithmic Injustices: Towards a Relational Ethics</a>. We caught up with Abeba, whose aforementioned paper was the recipient of the Best Paper award at the most recent Black in AI Workshop at NeurIPS, to go in-depth on the paper and the thought process around AI ethics. In our conversation, we discuss the “harm of categorization”, and how the thinking around these categorizations should be discussed, how ML generally doesn’t account for the ethics of various scenarios and how relational ethics could solve this issue, her most recent paper “<a href= "https://arxiv.org/pdf/2001.05046.pdf">Robot Rights? Let’s Talk about Human Welfare Instead</a>,” and much more. Check out our complete write-up and resource page at twimlai.com/talk/348.

Today we’re joined by Abeba Birhane, PhD Student at University College Dublin and author of the recent paper Algorithmic Injustices: Towards a Relational Ethics. We caught up with Abeba, whose aforementioned paper was the recipient of the Best Paper award at the most recent Black in AI Workshop at NeurIPS, to go in-depth on the paper and the thought process around AI ethics. In our conversation, we discuss the “harm of categorization”, and how the thinking around these categorizations should be discussed, how ML generally doesn’t account for the ethics of various scenarios and how relational ethics could solve this issue, her most recent paper “Robot Rights? Let’s Talk about Human Welfare Instead,” and much more. Check out our complete write-up and resource page at twimlai.com/talk/348. 

Algorithmic Injustices and Relational Ethics with Abeba Birhane - #348

Today we’re excited to kick off our annual Black in AI Series joined by Nemo Semret, CTO of Gro Intelligence. Gro provides an agricultural data platform dedicated to improving global food security, focused on applying AI at macro scale. In our conversation with Nemo, we discuss Gro’s approach to data acquisition, how they apply machine learning to various problems, and their approach to modeling.    Check out the full interview and show notes at <a href= "https://twimlai.com/talk/347">twimlai.com/talk/347</a>.

Today we’re excited to kick off our annual Black in AI Series joined by Nemo Semret, CTO of Gro Intelligence. Gro provides an agricultural data platform dedicated to improving global food security, focused on applying AI at macro scale. In our conversation with Nemo, we discuss Gro’s approach to data acquisition, how they apply machine learning to various problems, and their approach to modeling.    Check out the full interview and show notes at twimlai.com/talk/347.

AI for Agriculture and Global Food Security with Nemo Semret - #347

Today we’re joined by Ryan Rogers, Senior Software Engineer at LinkedIn. We caught up with Ryan at NeurIPS, where he presented the paper “Practical Differentially Private Top-k Selection with Pay-what-you-get Composition” as a spotlight talk. In our conversation, we discuss how LinkedIn allows its data scientists to access aggregate user data for exploratory analytics while maintaining its users’ privacy with differential privacy, and the major components of the paper. We also talk through one of the big innovations in the paper, which is discovering the connection between a common algorithm for implementing differential privacy, the exponential mechanism, and Gumbel noise, which is commonly used in machine learning.   The complete show notes for this episode can be found at <a href= "https://twimlai.com/talk/346">twimlai.com/talk/346</a>.

Today we’re joined by Ryan Rogers, Senior Software Engineer at LinkedIn. We caught up with Ryan at NeurIPS, where he presented the paper “Practical Differentially Private Top-k Selection with Pay-what-you-get Composition” as a spotlight talk. In our conversation, we discuss how LinkedIn allows its data scientists to access aggregate user data for exploratory analytics while maintaining its users’ privacy with differential privacy, and the major components of the paper. We also talk through one of the big innovations in the paper, which is discovering the connection between a common algorithm for implementing differential privacy, the exponential mechanism, and Gumbel noise, which is commonly used in machine learning.   The complete show notes for this episode can be found at twimlai.com/talk/346. 

Practical Differential Privacy at LinkedIn with Ryan Rogers - #346

Today we conclude our KubeCon ‘19 Series joined by Erez Cohen, VP of CloudX & AI at Mellanox. In our conversation, we discuss: <ul> <li style="font-weight: 400;">Erez’s talk “Networking Optimizations for Multi-Node Deep Learning on Kubernetes.” where he discusses problems and solutions related to networking discovered during the journey to reduce training time. </li> <li style="font-weight: 400;">NVIDIA’s recent acquisition of Mellanox, and what fruits that relationship hopes to bear. </li> <li style="font-weight: 400;">The evolution of technologies like RDMA, GPU Direct, and Sharp, Mellanox’s solution to improve the performance of MPI operations, which can be found in NVIDIA’s NCCL collective communications library.</li> <li style="font-weight: 400;">How Mellanox is enabling Kubernetes and other platforms to take advantage of the various technologies mentioned above. </li> <li style="font-weight: 400;">Why we should care about networking in Deep Learning, which is inherently a compute-bound process. </li> </ul> The complete show notes for this episode can be found at <a href= "https://twimlai.com/talk/345">twimlai.com/talk/345</a>.

Today we conclude our KubeCon ‘19 Series joined by Erez Cohen, VP of CloudX & AI at Mellanox. In our conversation, we discuss: Erez’s talk “Networking Optimizations for Multi-Node Deep Learning on Kubernetes.” where he discusses problems and solutions related to networking discovered during the journey to reduce training time.  NVIDIA’s recent acquisition of Mellanox, and what fruits that relationship hopes to bear.  The evolution of technologies like RDMA, GPU Direct, and Sharp, Mellanox’s solution to improve the performance of MPI operations, which can be found in NVIDIA’s NCCL collective communications library. How Mellanox is enabling Kubernetes and other platforms to take advantage of the various technologies mentioned above.  Why we should care about networking in Deep Learning, which is inherently a compute-bound process.  The complete show notes for this episode can be found at twimlai.com/talk/345.

Networking Optimizations for Multi-Node Deep Learning on Kubernetes with Erez Cohen - #345

Today we’re joined by Bob Killen, Research Cloud Administrator at the University of Michigan. In our conversation, we discuss: <ul> <li style="font-weight: 400;">How his group is deploying Kubernetes at UM.</li> <li style="font-weight: 400;">The user experience of his broad user base, including those using KubeFlow environments.</li> <li style="font-weight: 400;">How users are taking advantage of distributed computing.</li> <li style="font-weight: 400;">Should ML/AI focused Kubernetes users should fear that the larger non-ML/AI user base will negatively impact their feature needs?</li> <li style="font-weight: 400;">Where do the largest gaps currently exist in trying to support ML/AI users’ workloads?</li> <li style="font-weight: 400;">Where Bob sees things going from a user perspective, and what are the things those users are asking about most? </li> </ul> The complete show notes for this episode can be found at <a href= "https://twimlai.com/talk/344">twimlai.com/talk/344</a>.

Today we’re joined by Bob Killen, Research Cloud Administrator at the University of Michigan. In our conversation, we discuss: How his group is deploying Kubernetes at UM. The user experience of his broad user base, including those using KubeFlow environments. How users are taking advantage of distributed computing. Should ML/AI focused Kubernetes users should fear that the larger non-ML/AI user base will negatively impact their feature needs? Where do the largest gaps currently exist in trying to support ML/AI users’ workloads? Where Bob sees things going from a user perspective, and what are the things those users are asking about most?  The complete show notes for this episode can be found at twimlai.com/talk/344.

Managing Research Needs at the University of Michigan using Kubernetes w/ Bob Killen - #344

This Week in Machine Learning & AI is the most popular podcast of its kind. TWiML & AI caters to a highly-targeted audience of machine learning & AI enthusiasts. They are data scientists, developers, founders, CTOs, engineers, architects, IT & product leaders, as well as tech-savvy business leaders. These creators, builders, makers and influencers value TWiML as an authentic, trusted and insightful guide to all that’s interesting and important in the world of machine learning and AI. Technologies covered include: machine learning, artificial intelligence, deep learning, natural language processing, neural networks, analytics, deep learning and more.

hubhopper

This Week in Machine Learning & Artificial Intelligence (AI)...

hubhopper