Richard Socher — The Challenges of Making ML Work in the Real World

09/29/20 • 50 min

Richard Socher, ex-Chief Scientist at Salesforce, joins us to talk about The AI Economist, NLP protein generation and biggest challenge in making ML work in the real world. Richard Socher was the Chief scientist (EVP) at Salesforce where he lead teams working on fundamental research(einstein.ai/), applied research, product incubation, CRM search, customer service automation and a cross-product AI platform for unstructured and structured data. Previously, he was an adjunct professor at Stanford’s computer science department and the founder and CEO/CTO of MetaMind(www.metamind.io/) which was acquired by Salesforce in 2016. In 2014, he got my PhD in the [CS Department](www.cs.stanford.edu/) at Stanford. He likes paramotoring and water adventures, traveling and photography. More info: - Forbes article: https://www.forbes.com/sites/gilpress/2017/05/01/emerging-artificial-intelligence-ai-leaders-richard-socher-salesforce/) with more info about Richard's bio. - CS224n - NLP with Deep Learning(http://cs224n.stanford.edu/) the class Richard used to teach. - TEDx talk(https://www.youtube.com/watch?v=8cmx7V4oIR8) about where AI is today and where it's going. Research: Google Scholar Link(https://scholar.google.com/citations?user=FaOcyfMAAAAJ&hl=en) The AI Economist: Improving Equality and Productivity with AI-Driven Tax Policies Arxiv link(https://arxiv.org/abs/2004.13332), blog(https://blog.einstein.ai/the-ai-economist/), short video(https://www.youtube.com/watch?v=4iQUcGyQhdA), Q&A(https://salesforce.com/company/news-press/stories/2020/4/salesforce-ai-economist/), Press: VentureBeat(https://venturebeat.com/2020/04/29/salesforces-ai-economist-taps-reinforcement-learning-to-generate-optimal-tax-policies/), TechCrunch(https://techcrunch.com/2020/04/29/salesforce-researchers-are-working-on-an-ai-economist-for-more-equitable-tax-policy/) ProGen: Language Modeling for Protein Generation: bioRxiv link(https://www.biorxiv.org/content/10.1101/2020.03.07.982272v2), [blog](https://blog.einstein.ai/progen/) ] Dye-sensitized solar cells under ambient light powering machine learning: towards autonomous smart sensors for the internet of things Issue11, (**Chemical Science 2020**). paper link(https://pubs.rsc.org/en/content/articlelanding/2020/sc/c9sc06145b#!divAbstract) CTRL: A Conditional Transformer Language Model for Controllable Generation: Arxiv link(https://arxiv.org/abs/1909.05858), code pre-trained and fine-tuning(https://github.com/salesforce/ctrl), blog(https://blog.einstein.ai/introducing-a-conditional-transformer-language-model-for-controllable-generation/) Genie: a generator of natural language semantic parsers for virtual assistant commands: PLDI 2019 pdf link(https://almond-static.stanford.edu/papers/genie-pldi19.pdf), https://almond.stanford.edu Topics Covered: 0:00 intro 0:42 the AI economist 7:08 the objective function and Gini Coefficient 12:13 on growing up in Eastern Germany and cultural differences 15:02 Language models for protein generation (ProGen) 27:53 CTRL: conditional transformer language model for controllable generation 37:52 Businesses vs Academia 40:00 What ML applications are important to salesforce 44:57 an underrated aspect of machine learning 48:13 Biggest challenge in making ML work in the real world Visit our podcasts homepage for transcripts and more episodes! www.wandb.com/podcast Get our podcast on Soundcloud, Apple, Spotify, and Google! Soundcloud: https://bit.ly/2YnGjIq Apple Podcasts: https://bit.ly/2WdrUvI Spotify: https://bit.ly/2SqtadF Google: http://tiny.cc/GD_Google Weights and Biases makes developer tools for deep learning. Join our bi-weekly virtual salon and listen to industry leaders and researchers in machine learning share their research: http://tiny.cc/wb-salon Join our community of ML practitioners: http://bit.ly/wb-slack Our gallery features curated machine learning reports by ML researchers. https://app.wandb.ai/gallery

Previous Episode

Zack Chase Lipton — The Medical Machine Learning Landscape

How Zack went from being a musician to professor, how medical applications of Machine Learning are developing, and the challenges of counteracting bias in real world applications. Zachary Chase Lipton is an assistant professor of Operations Research and Machine Learning at Carnegie Mellon University. His research spans core machine learning methods and their social impact and addresses diverse application areas, including clinical medicine and natural language processing. Current research focuses include robustness under distribution shift, breast cancer screening, the effective and equitable allocation of organs, and the intersection of causal thinking with messy data. He is the founder of the Approximately Correct (approximatelycorrect.com) blog and the creator of Dive Into Deep Learning, an interactive open-source book drafted entirely through Jupyter notebooks. Zack’s blog - http://approximatelycorrect.com/ Detecting and Correcting for Label Shift with Black Box Predictors: https://arxiv.org/pdf/1802.03916.pdf Algorithmic Fairness from a Non-Ideal Perspective https://www.datascience.columbia.edu/data-good-zachary-lipton-lecture Jonas Peter’s lectures on causality: https://youtu.be/zvrcyqcN9Wo 0:00 Sneak peek: Is this a problem worth solving? 0:38 Intro 1:23 Zack’s journey from being a musician to a professor at CMU 4:45 Applying machine learning to medical imaging 10:14 Exploring new frontiers: the most impressive deep learning applications for healthcare 12:45 Evaluating the models – Are they ready to be deployed in hospitals for use by doctors? 19:16 Capturing the signals in evolving representations of healthcare data 27:00 How does the data we capture affect the predictions we make 30:40 Distinguishing between associations and correlations in data – Horror vs romance movies 34:20 The positive effects of augmenting datasets with counterfactually flipped data 39:25 Algorithmic fairness in the real world 41:03 What does it mean to say your model isn’t biased? 43:40 Real world implications of decisions to counteract model bias 49:10 The pragmatic approach to counteracting bias in a non-ideal world 51:24 An underrated aspect of machine learning 55:11 Why defining the problem is the biggest challenge for machine learning in the real world Visit our podcasts homepage for transcripts and more episodes! www.wandb.com/podcast Get our podcast on YouTube, Apple, and Spotify! YouTube: https://www.youtube.com/c/WeightsBiases Soundcloud: https://bit.ly/2YnGjIq Apple Podcasts: https://bit.ly/2WdrUvI Spotify: https://bit.ly/2SqtadF We started Weights and Biases to build tools for Machine Learning practitioners because we care a lot about the impact that Machine Learning can have in the world and we love working in the trenches with the people building these models. One of the most fun things about these building tools has been the conversations with these ML practitioners and learning about the interesting things they’re working on. This process has been so fun that we wanted to open it up to the world in the form of our new podcast called Gradient Dissent. We hope you have as much fun listening to it as we had making it! Join our bi-weekly virtual salon and listen to industry leaders and researchers in machine learning share their research: http://tiny.cc/wb-salon Join our community of ML practitioners where we host AMA's, share interesting projects and meet other people working in Deep Learning: http://bit.ly/wandb-forum Our gallery features curated machine learning reports by researchers exploring deep learning techniques, Kagglers showcasing winning models, and industry leaders sharing best practices. https://app.wandb.ai/gallery

Next Episode

Joaquin Candela — Definitions of Fairness

Joaquin chats about scaling and democratizing AI at Facebook, while understanding fairness and algorithmic bias. --- Joaquin Quiñonero Candela is Distinguished Tech Lead for Responsible AI at Facebook, where he aims to understand and mitigate the risks and unintended consequences of the widespread use of AI across Facebook. He was previously Director of Society and AI Lab and Director of Engineering for Applied ML. Before joining Facebook, Joaquin taught at the University of Cambridge, and worked at Microsoft Research. Connect with Joaquin: Personal website: https://quinonero.net/ Twitter: https://twitter.com/jquinonero LinkedIn: https://www.linkedin.com/in/joaquin-qui%C3%B1onero-candela-440844/ --- Topics Discussed: 0:00 Intro, sneak peak 0:53 Looking back at building and scaling AI at Facebook 10:31 How do you ship a model every week? 15:36 Getting buy-in to use a system 19:36 More on ML tools 24:01 Responsible AI at Facebook 38:33 How to engage with those effected by ML decisions 41:54 Approaches to fairness 53:10 How to know things are built right 59:34 Diversity, inclusion, and AI 1:14:21 Underrated aspect of AI 1:16:43 Hardest thing when putting models into production Transcript: http://wandb.me/gd-joaquin-candela Links Discussed: Race and Gender (2019): https://arxiv.org/pdf/1908.06165.pdf Lessons from Archives: Strategies for Collecting Sociocultural Data in Machine Learning (2019): https://arxiv.org/abs/1912.10389 Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification (2018): http://proceedings.mlr.press/v81/buolamwini18a.html --- Get our podcast on these platforms: Apple Podcasts: http://wandb.me/apple-podcasts Spotify: http://wandb.me/spotify Google Podcasts: http://wandb.me/google-podcasts YouTube: http://wandb.me/youtube Soundcloud: http://wandb.me/soundcloud Join our community of ML practitioners where we host AMAs, share interesting projects and meet other people working in Deep Learning: http://wandb.me/slack Check out Fully Connected, which features curated machine learning reports by researchers exploring deep learning techniques, Kagglers showcasing winning models, industry leaders sharing best practices, and more: https://wandb.ai/fully-connected