Schedule

Abstract: Deep learning approaches are increasingly used in healthcare due to their seemingly remarkable performance. However, they can be notoriously brittle, often with little ability to generalize outside their training data. Using real life examples from ophthalmology, oncology and radiology, we will first discuss practical examples of distribution shifts. We will then highlight how even seemingly subtle distribution shifts can lead to catastrophic failures of models. We will highlight the need for constant vigilance of the input data and better metrics to quantify distribution shifts. We will conclude with a plea to the ICML/PODS community to work with clinical community on this critically important topic.

Abstract: Machine learning models deployed in the real world constantly face distribution shifts, and these distribution shifts can significantly degrade model performance. In this talk, I will present the WILDS benchmark of real-world distribution shifts, focusing on the version 2.0 update that adds curated unlabeled data. Unlabeled data can be a powerful leverage for improving out-of-distribution performance, but existing distribution shift benchmarks with unlabeled data do not reflect the breadth of scenarios that arise in real-world applications. To this end, we provide unlabeled data to 8 out of 10 datasets in WILDS, spanning diverse applications and modalities. We observe that existing methods fail to improve out-of-distribution performance on WILDS, even though these methods have been successful on existing benchmarks with different types of distribution shifts. This underscores the importance of developing and evaluating methods on diverse types of distribution shifts, including directly on shifts that arise in practice.

Abstract: Machine learning models often perform poorly under distribution shift. But can we understand how a particular distribution shift will affect a model? We approach this in two parts: (1) explaining the shift itself, and (2) explaining the model's behavior.

First, we train a language model to describe the difference between two distributions. The model produces natural language explanations that allow humans to distinguish random draws from the two distributions. This helps reveal subtle but important shifts that may not be apparent from manual inspection, and can also be used to uncover spurious cues. We use this to identify "shortcuts" that models rely on, and construct a distribution shift that breaks the shortcut and decreases model performance.

Having built tools to understand *how* the data is shifted, we next investigate whether model explanations (such as Grad-CAM) can be used to predict the behavior of models under distribution shift. Here, the resuts are largely negative. We construct models with specific defects (such as backdoors or spurious cues) that affect out-of-distribution performance, and measure whether model explanations can distinguish these from regular, non-defective models. Detection rates are typically low and in some cases trivial. This underscores the need to improve model explanations if they are to be used as a reliable tool for model debugging.

Abstract: Given the inherent difficulty of learning a model that is robust to data distribution shifts, much research focus has `shifted’ to learning data representations that are useful for learning good models for downstream, yet unknown, data distributions.

The primary aim of such models is accuracy generalization.

In this talk I wish to address an additional desideratum—model fairness.

On a high level, the question I am interested in is: to what extent and under what assumptions can one come up with data representations that are both “fair” and allow accurate predictions when applied to downstream tasks about which one has only limited information?

I will address different possible fairness requirements and provide some initial insights on what can, and more often, what cannot be achieved along this line.

Abstract: The formalism of structural causal models provides a precise approach for describing certain types of distribution shifts, via the notion of a soft intervention or mechanism change. Popular approaches to learning causal structure from data rely on the availability of distribution shifts in order to identify between otherwise indistinguishable models. However, many approaches rely on prior knowledge of which variables have been shifted between settings, called intervention targets. When this information is not available, one must simultaneously learn both intervention targets and the causal structure. We introduce the Unknown-Target Interventional Greedy Sparsest Permutation algorithm, a nonparametric, hybrid approach for this learning task. We prove the consistency of the algorithm, and demonstrate its performance on synthetic and biological datasets.

Abstract: Over the past few years, there has been exciting progress on algorithmic robust statistics in unsupervised, supervised and online learning. Much of this progress has been fueled by new algorithmic tools for detecting portions of the samples that have different distributional profiles. We will survey some of these tools as well as discuss prospects for building theories of coping with distribution shift from them.

Abstract: Growing interest in the external validity of prediction models has produced many methods for finding predictive distributions that are invariant to dataset shifts and can be used for prediction in new, unseen environments. However, these methods consider different types of shifts and have been developed under disparate frameworks, making it difficult to theoretically analyze how solutions differ with respect to stability and accuracy. Taking a causal graphical view, in this talk I will discuss three graphical operators for removing unstable parts of the DGP that correspond to three types of stable distributions. This clarifies the relationship between the types of “invariances” sought by many existing methods. Then, using an example from healthcare, I will demonstrate the tradeoff between minimax and average performance, highlighting the need for model developers to carefully determine when and how they achieve invariance.

Léon received the Diplôme d'Ingénieur de l'École Polytechnique (X84) in 1987, the Magistère de Mathématiques Fondamentales et Appliquées et d'Informatique from École Normale Superieure in 1988, the Diplôme d'Études Approndies in Computer Science in 1988, and a Ph.D. in Computer Science from LRI, Université de Paris-Sud in 1991.

After his Ph.D., Léon went to AT&T Bell Laboratories from 1991 to 1992. He then became chairman of Neuristique, a small company pioneering machine learning for data mining applications. He returned to AT&T Labs from 1995 to 2002, NEC Labs America at Princeton from 2002 to 2010, and Microsoft from 2010 to February 2015. He joined the Facebook AI Research in March 2015.

Léon's primary research interest is machine learning. His contributions to this field address theory, algorithms and large scale applications. Léon's secondary research interest is data compression and coding. His best known contributions are his work on large scale learning and on the DjVu document compression technology. Léon has published about one hundred scientific papers. He is serving or has served on the boards of the Journal of Machine Learning Research, IEEE Transactions on Pattern Analysis and Machine and Pattern Recognition Letters. He won the 2007 Blavatnik Award for young scientists.

Dr. Kalpathy-Cramer is an Associate Professor of Radiology at Harvard Medical School, Co-Director of the QTIM lab and the Center for Machine Learning at the Athinoula A. Martinos Center and Scientific Director at the MGH & BWH Center for Clinical Data Science. Her research areas include machine learning, informatics, image analysis and statistical methods. In addition to developing novel machine learning algorithms, her lab is also actively engaged in the applications of these to clinical problems in radiology, oncology and ophthalmology.

She is funded through NIH to develop quantitative imaging methods in cancer. She is the PI of an NSF-funded project to develop novel algorithms and apply them to build diagnostic tools in ophthalmology. Research from this work has resulted in a deep-learning based algorithm for disease diagnosis and response assessment that is currently being evaluated at several clinics and screening trials in the US and India. Her group has recently applied novel machine learning methods to stroke segmentation, identification, and outcome prediction. She leads an effort to develop open-source tools for deep learning based image analysis and is making trained models for brain tumor, stroke and other diseases publicly available.

Percy Liang is an Associate Professor of Computer Science at Stanford University (B.S. from MIT, 2004; Ph.D. from UC Berkeley, 2011). His two research goals are (i) to make machine learning more robust, fair, and interpretable; and (ii) to make computers easier to communicate with through natural language. His awards include the Presidential Early Career Award for Scientists and Engineers (2019), IJCAI Computers and Thought Award (2016), an NSF CAREER Award (2016), a Sloan Research Fellowship (2015), and a Microsoft Research Faculty Fellowship (2014).

Alexander J. Smola studied physics in Munich at the University of Technology, Munich, at the Universita degli Studi di Pavia and at AT&T Research in Holmdel. During this time he was at the Maximilianeum München and the Collegio Ghislieri in Pavia. In 1996 he received the Master degree at the University of Technology, Munich and in 1998 the Doctoral Degree in computer science at the University of Technology Berlin. Until 1999 he was a researcher at the IDA Group of the GMD Institute for Software Engineering and Computer Architecture in Berlin (now part of the Fraunhofer Geselschaft). After that, he worked as a Researcher and Group Leader at the Research School for Information Sciences and Engineering of the Australian National University. From 2004 onwards he worked as a Senior Principal Researcher and Program Leader at the Statistical Machine Learning Program at NICTA. From 2008 to 2012 he worked at Yahoo Research. In spring of 2012 he moved to Google Research to spend a wonderful year in Mountain View and he continued working there until the end of 2014. From 2013-2017 he was professor at Carnegie Mellon University. He co-founded Marianas Labs in early 2015. In July 2016 he moved to Amazon Web Services to help build AI and Machine Learning tools for everyone.