Xenobots and Critical Learning Periods

This week we take a look at the need for pooling layers within CNNs as well as discussing the regularization of CNNs using large-scale neuroscience data.We are also very pleased to have Rosanne Liu join us on the show. Rosanne is a senior research scientist and a founding member of Uber AI. She is interested in making neural networks a better place and also currently runs a deep learning reading group called "Deep Learning: Classics and Trends".Rosanne Liu Twitter: https://twitter.com/savvyrl?lang=enPlease let us know who you thought presented the most underrated paper in the form below:https://forms.gle/97MgHvTkXgdB41TC8Also let us know any suggestions for future papers or guests:https://docs.google.com/forms/d/e/1FAIpQLSeWoZnImRHXy8MTeBhKA4bxRPVVnVXAUb0bLIP0bQpiTwX6uA/viewformLinks to the papers:"Learning From Brains How to Regularize Machines" - https://arxiv.org/pdf/1911.05072.pdf"Pooling is neither necessary nor sufficient for appropriate deformation stability in CNNs - https://arxiv.org/pdf/1804.04438.pdf"Plug and play language models: A simple approach to controlled text generation" - https://arxiv.org/pdf/1912.02164.pdf

We open season two of Underrated ML with Anna Huang on the show. Anna Huang is a Research Scientist at Google Brain, working on the Magenta project. Her research focuses on designing generative models to make creating music more approachable. She is the creator of Music Transformer and also the ML model Coconet that powered Google’s first AI Doodle the Bach Doodle.She holds a PhD in computer science from Harvard University and was a recipient of the NSF Graduate Research Fellowship. She spent the later parts of her PhD as a visiting research student at the Montreal Institute of Learning Algorithms (MILA). She publishes in machine learning, human-computer interaction, and music, at conferences such as ICLR, IUI, CHI, and ISMIR.She has been a judge on the Eurovision AI Song Contest and her compositions have won awards including first place in the San Francisco Choral Artists’ a cappella composition contest. She holds a masters in media arts and sciences from the MIT Media Lab, and a B.S. in computer science and B.M. in music composition both from the University of Southern California. She grew up in Hong Kong, where she learned to play the guzheng.On the episode we discuss Metaphoria by Kate Gero and Lydia Chilton, which is a fascinating tool allowing users to generate metaphors from only a select number of words. We also discuss the current trends regarding the dangers of AI with a case study on child welfare.Underrated ML Twitter: https://twitter.com/underrated_mlAnna Huang Twitter: https://twitter.com/huangczaPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:Gero, Katy Ilonka, and Lydia B. Chilton. "Metaphoria: An Algorithmic Companion for Metaphor Creation." CHI 2019. [paper][online paper] [talk] [demo]"A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions" - [paper]Additional Links:Compton, Kate, and Michael Mateas. "Casual Creators." ICCC 2015. [paper]Fiebrink, Rebecca, Dan Trueman, and Perry R. Cook. "A Meta-Instrument for Interactive, On-the-Fly Machine Learning." NIME 2009. [paper][talk][tool]Huang, Cheng-Zhi Anna, et al. "The Bach Doodle: Approachable music composition with machine learning at scale." ISMIR 2019. [paper][blog][doodle]

This week we are joined by Kyunghyun Cho. He is an associate professor of computer science and data science at New York University, a research scientist at Facebook AI Research and a CIFAR Associate Fellow. On top of this he also co-chaired the recent ICLR 2020 virtual conference.We talk about a variety of topics in this weeks episode including the recent ICLR conference, energy functions, shortcut learning and the roles popularized Deep Learning research areas play in answering the question “What is Intelligence?”.Underrated ML Twitter: https://twitter.com/underrated_mlKyunghyun Cho Twitter: https://twitter.com/kchonyc?ref_src=twsrc%5Egoogle%7Ctwcamp%5Eserp%7Ctwgr%5EauthorPlease let us know who you thought presented the most underrated paper in the form below:https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:“Shortcut Learning in Deep Neural Networks” - https://arxiv.org/pdf/2004.07780.pdf"Bayesian Deep Learning and a Probabilistic Perspective of Generalization” - https://arxiv.org/abs/2002.08791"Classifier-agnostic saliency map extraction" - https://arxiv.org/abs/1805.08249“Deep Energy Estimator Networks” - https://arxiv.org/abs/1805.08306“End-to-End Learning for Structured Prediction Energy Networks” - https://arxiv.org/abs/1703.05667“On approximating nabla f with neural networks” - https://arxiv.org/abs/1910.12744“Adversarial NLI: A New Benchmark for Natural Language Understanding“ - https://arxiv.org/abs/1910.14599“Learning the Difference that Makes a Difference with Counterfactually-Augmented Data” - https://arxiv.org/abs/1909.12434“Learning Concepts with Energy Functions” - https://openai.com/blog/learning-concepts-with-energy-functions/

This week we are joined by Julius Adebayo. Julius is a CS PhD student at MIT, interested in safe deployment of ML based systems as it relates to privacy/security, interpretability, fairness and robustness.He is motivated by the need to ensure that ML based systems demonstrate safe behaviour when deployed.On this weeks episode we discuss how the evolution of hardware has progressed overtime and what that means for deep learning research. We also analyse how microprocessors can aid developments in neuroscience understanding.Underrated ML Twitter: https://twitter.com/underrated_mlJulius Adebayo Twitter: https://twitter.com/julius_adebayoPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:"Could a Neuroscientist Understand a Microprocessor?" [paper]"When will computer hardware match the human brain?" - [paper]

This week we are joined by Ari Morcos. Ari is a research scientist at Facebook AI Research (FAIR) in Menlo Park working on understanding the mechanisms underlying neural network computation and function, and using these insights to build machine learning systems more intelligently. In particular, he has worked on a variety of topics, including understanding the lottery ticket hypothesis, self-supervised learning, the mechanisms underlying common regularizers, and the properties predictive of generalization, as well as methods to compare representations across networks, the role of single units in computation, and on strategies to measure abstraction in neural network representations. Previously, he worked at DeepMind in London.Ari earned his PhD working with Chris Harvey at Harvard University. For his thesis, he developed methods to understand how neuronal circuits perform the computations necessary for complex behaviour. In particular, his research focused on how parietal cortex contributes to evidence accumulation decision-making.In this episode, we discuss the importance of certain layers within neural networks.Underrated ML Twitter: https://twitter.com/underrated_mlNaila Murray Twitter: https://twitter.com/arimorcosPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Link to the paper:"Are All Layers Created Equal?" [paper]

This week we are joined by Sebastian Ruder. He is a research scientist at DeepMind, London. He has also worked at a variety of institutions such as AYLIEN, Microsoft, IBM's Extreme Blue, Google Summer of Code, and SAP. These experiences were completed in tangent with his studies which included studying Computational Linguistics at the University of Heidelberg, Germany and at Trinity College, Dublin before undertaking a PhD in Natural Language Processing and Deep Learning at the Insight Research Centre for Data Analytics.This week we discuss language independence and diversity in natural language processing whilst also taking a look at the attempts to identify material properties from images.As discussed in the podcast if you would like to donate to the current campaign of "CREATE DONATE EDUCATE" which supports Stop Hate UK then please find the link below:https://www.shorturl.at/glmszPlease also find additional links to help support black colleagues in the area of research;Black in AI twitter account: https://twitter.com/black_in_aiMentoring and proofreading sign-up to support our Black colleagues in research: https://twitter.com/le_roux_nicolas/status/1267896907621433344?s=20Underrated ML Twitter: https://twitter.com/underrated_mlSebastian Ruder Twitter: https://twitter.com/seb_ruderPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:“On Achieving and Evaluating Language-Independence in NLP” - https://journals.linguisticsociety.org/elanguage/lilt/article/view/2624.html"The State and Fate of Linguistic Diversity and Inclusion in the NLP World” - https://arxiv.org/abs/2004.09095"Recognizing Material Properties from Images" - https://arxiv.org/pdf/1801.03127.pdfAdditional Links:Student perspectives on applying to NLP PhD programs: https://blog.nelsonliu.me/2019/10/24/student-perspectives-on-applying-to-nlp-phd-programs/Tim Dettmer's post on how to pick your grad school: https://timdettmers.com/2020/03/10/how-to-pick-your-grad-school/Rachel Thomas' blog post on why you should blog: https://medium.com/@racheltho/why-you-yes-you-should-blog-7d2544ac1045Emily Bender's The Gradient article: https://thegradient.pub/the-benderrule-on-naming-the-languages-we-study-and-why-it-matters/Paper on order-sensitive vs order-free methods: https://www.aclweb.org/anthology/N19-1253.pdf"Exploring the Origins and Prevalence of Texture Bias in Convolutional Neural Networks": https://arxiv.org/abs/1911.09071Sebastian's website where you can find all his blog posts: https://ruder.io/

This week we are joined by Naila Murray. Naila obtained a B.Sc. in Electrical Engineering from Princeton University in 2007. In 2012, she received her PhD from the Universitat Autonoma de Barcelona, in affiliation with the Computer Vision Center. She joined NAVER LABS Europe (then Xerox Research Centre Europe) in January 2013, working on topics including fine-grained visual categorization, image retrieval, and visual attention. From 2015 to 2019 she led the computer vision team at NLE. She currently serves as NLE's director of science. She serves/served as area chair for ICLR 2018, ICCV 2019, ICLR 2019, CVPR 2020, ECCV 2020, and programme chair for ICLR 2021. Her research interests include representation learning and multi-modal search.We discuss using sparse pairwise comparisons to learn a ranking function that is robust to outliers. We also take a look at using generative models in order to utilise once inaccessible datasets.Underrated ML Twitter: https://twitter.com/underrated_mlNaila Murray Twitter: https://twitter.com/NailaMurrayPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:"Interestingness Prediction by Robust Learning to Rank" [paper]"Generative Models for Effective ML on Private Decentralized datasets" - [paper]

We conclude season one of Underrated ML by having Stephen Merity on as our guest. Stephen has worked at various institutions such as MetaMind and Salesforce ohana, Google Sydney, Freelancer.com, the Schwa Lab at the University of Sydney, the team at Grok Learning, the non-profit Common Crawl, and IACS @ Harvard. He also holds a Bachelor of Information Technology from the University of Sydney and a Master of Science in Computational Science and Engineering from Harvard University.In this weeks episode we talk about the current influences of hardware in the field of Deep Learning research, baseline models, strongly typed RNNs and Alan Turings paper on the chemical basis of morphogenesis.Underrated ML Twitter: https://twitter.com/underrated_mlStephen Merity Twitter: https://twitter.com/SmerityPlease let us know who you thought presented the most underrated paper in the form below: https://forms.gle/97MgHvTkXgdB41TC8Links to the papers:“The Chemical Basis of Morphogenesis” - https://www.dna.caltech.edu/courses/cs191/paperscs191/turing.pdf"Strongly-Typed Recurrent Neural Networks” - https://arxiv.org/abs/1602.02218"Quasi-Recurrent Neural Networks" - https://arxiv.org/abs/1611.01576"An Analysis of Neural Language Modelling at Multiple Scales" - https://arxiv.org/abs/1803.08240Additional Links:Aleatory architecture / hysteresis: Why Birds Are The World's Best EngineersNear decomposability: Near decomposability and the speed of evolution / The Architecture of ComplexityGoogle's All Our N-gram are Belong to You from 2006

This weeks episode we take a look at Abstract Reasoning within Neural Networks as well as discussing the current review system surrounding ML papers. We are also very happy to have Jacob Buckman join us on the podcast this week. Jacob is currently undertaking a PhD at Mila having previously been a researcher at Google Brain with Sara Hooker. His main research interests lie in deep reinforcement learning with a particular focus on sample-efficiency. Please let us know who you thought presented the most underrated paper in the form below:https://forms.gle/97MgHvTkXgdB41TC8Also let us know any suggestions for future papers or guests:https://docs.google.com/forms/d/e/1FAIpQLSeWoZnImRHXy8MTeBhKA4bxRPVVnVXAUb0bLIP0bQpiTwX6uA/viewformLinks to the papers:"Conference Reviewing Considered Harmful" - http://pages.cs.wisc.edu/~dusseau/Classes/CS739/anderson-model.pdf"Measuring Abstract Reasoning in Neural Networks" - http://proceedings.mlr.press/v80/santoro18a/santoro18a.pdf?fbclid=IwAR2rqCYu_rorfiVicYXx4EnGFZ4Y-9uAh9936YxEEwGxY-5MGGbnm9CMfXI

Have a listen to the first ever Underrated ML podcast! We'll walk you through two papers which we found really interesting followed by a few questions and then finally finishing with our verdict on what we believe was the most underrated paper!Links to the papers can be found below.Critical Learning Periods in Deep Neural Networks - https://arxiv.org/abs/1711.08856A scalable pipeline for designing reconfigurable organisms - https://www.pnas.org/content/117/4/1853