Researchers Comparing DNNs to Brains Need to Adopt Standard Methods of Science.

The claim that DNNs and brains represent information in similar ways is largely based on the good performance of DNNs on various brain benchmarks. On this approach, the better DNNs can predict neural activity, the better the correspondence between DNNs and brains. But this is at odds with the standard scientific research approach that is characterized by varying independent variables to test specific hypotheses regarding the causal mechanisms that underlie some phenomenon; models are supported to the extent that they account for these experimental results. The best evidence for a model is that it survives “severe” tests, namely, experiments that have a high probability of falsifying a model if and only if the model is false in some relevant manner. When DNNs are assessed in this way, they catastrophically fail. The field needs to change its methods and put far more weight into falsification to get a better characterization of DNN-brain correspondences and to build more human-like AI.

Bio

Jeffrey Bowers received his PhD in psychology at the University of Arizona, was an assistant professor at Rice University in Texas, and has been at University of Bristol in the UK since 1998 where he is now a full professor of psychology. Jeffrey has carried out empirical work concerning memory, language and memory, and more recently, has been exploring the links between DNNs and psychology, particularly in the domain of vision.

On the Elements of Theory in Neuroscience.

In science, theories are essential for encapsulating knowledge obtained from data, making predictions, and building models that make simulations and technological applications possible. Neuroscience -- along with cognitive science -- however, is a young field with fewer established theories (than, say, physics). One consequence of this fact is that new practitioners in the field sometimes find it difficult to know what makes a good theory. Moreover, the use of conceptual theories and models in the field has endured some criticisms: theories have low quantitative prediction power; models have weak transparency; etc. Addressing these issues calls for identifying the elements of theory in neuroscience. In this talk I will try to present and discuss, with case studies, the following: (1) taxonomies by which the different dimensions of a theory can be assessed. (2) criteria for the goodness of a theory. (3)trade-offs between agreement with the natural world and representational consistency in the theory/model world.

Bio

Dr. Lawrence Udeigwe is an Associate Professor of Mathematics at Manhattan College and a 2021/22 MLK Visiting Associate Professor in Brain and Cognitive Sciences at MIT. His research interests include: use of differential equations to understand the dynamical interaction between Hebbian plasticity and homeostatic plasticity; use of artificial neural networks (ANN) to investigate the mechanisms behind surround suppression and other vision normalization processes; and exploring the practical and philosophical implications of the use of theory in neuroscience. Dr. Udeigwe obtained his PhD from the University of Pittsburgh in 2014 under the supervision of Bard Ermentrout and Paul Munro.

On the Limitations of Bayesian Uncertainty in Adversarial Settings.

Adversarial examples have been recognized as a threat, and still pose problems, as it is hard to defend them. Naturally, one might be tempted to think that an image looking like a panda and being classified as a gibbon might be unusual-or at least unusual enough to be discovered by for example Bayesian uncertainty measures. Alas, it turns out that also Bayesian confidence and uncertainty measures are easy to fool when the optimization procedure is adapted accordingly. Moreover, adversarial examples transfer between different methods, so they can also be attacked in a black box setting. To conclude the talk, we will discuss briefly the practical necessity to defend evasion, and what is needed to not only evaluate defenses properly, but also build practical defenses.

Bio

Kathrin Grosse is a Postdoctoral Researcher at the PRA Lab at the University of Cagliari, Italy. She received her MSc in 2016 and her PhD in 2021 from Saarland university, under the supervision of Michael Backes at the CISPA Helmholtz Center. Her research interests are the intersection of machine learning and security, and recently focus on training time attacks and the study of machine learning security in practice. During her PhD, she interned at Disney Research Zurich in 2020/21 and IBM Yorktown in 2019. Furthermore, she was nominated as an AI Newcommer in the context of the German Federal Ministry of Education and Research’s Science Year 2019. She serves as a reviewer for many international journals and conferences.

Causal Principles Meet Deep Learning: Successes and Challenges.

This talk is concerned with causal representation learning, which aims to reveal the underlying high-level hidden causal variables and their relations. It can be seen as a special case of causal discovery, whose goal is to recover the underlying causal structure or causal model from observational data. The modularity property of a causal system implies properties of minimal changes and independent changes of causal representations, and I will explain how such properties make it possible to recover the underlying causal representations from observational data with identifiability guarantees: under appropriate assumptions, the learned representations are consistent with the underlying causal process. The talk will consider various settings with independent and identically distributed (i.i.d.) data, temporal data, or data with distribution shift as input, and demonstrate when identifiable causal representation learning can benefit from the flexibility of deep learning and when it has to impose parametric assumptions on the causal process.

Bio

Kun Zhang is an associate professor of philosophy and an affiliate faculty in the machine learning department of Carnegie Mellon University (CMU); he is currently on leave from CMU and working as an associate professor of machine learning at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI). He has been actively developing methods for automated causal discovery from various kinds of data and investigating machine learning problems including transfer learning, representation learning, and reinforcement learning from a causal perspective. He has been frequently serving as a senior area chair, area chair, or senior program committee member for major conferences in machine learning or artificial intelligence, including NeurIPS, ICML, UAI, IJCAI, AISTATS, and ICLR, and is a general & program co-chair of the first Conference on Causal Learning and Reasoning (CLeaR 2022) and a program co-chair of the 38th Conference on Uncertainty in Artificial Intelligence (UAI 2022).

Surprising failures of standard practices in ML when the sample size is small.

In this talk, we discuss two failure cases of common practices that are typically believed to improve on vanilla methods: (i) adversarial training can lead to worse robust accuracy than standard training (ii) active learning can lead to a worse classifier than a model trained using uniform samples. In particular, we can prove both mathematically and empirically, that such failures can happen in the small-sample regime. We discuss high-level explanations derived from the theory, that shed light on the causes of these phenomena in practice.

Bio

Fanny Yang is an Assistant Professor of Computer Science at ETH Zurich. She received her Ph.D. in EECS from University of California, Berkeley in 2018 and was a postdoctoral fellow at Stanford University and ETH-ITS in 2019. Her current research interests include methodological and theoretical advances for problems that arise from distribution shift or adversarial robustness requirements, studying the (robust) generalization ability of overparameterized models for high-dimensional data, and interpretability/explainability of neural networks.

When Bayesian Orthodoxy Can Go Wrong: Model Selection and Out-of-Distribution Generalization

We will re-examine two popular use-cases of Bayesian approaches: model selection, and robustness to distribution shifts.

The marginal likelihood (Bayesian evidence) provides a distinctive approach to resolving foundational scientific questions --- "how can we choose between models that are entirely consistent with any data?" and "how can we learn hyperparameters or correct ground truth constraints, such as intrinsic dimensionalities, or symmetries, if our training loss doesn't select for them?". There are compelling arguments that the marginal likelihood automatically encodes Occam's razor. There are also widespread practical applications, including the variational ELBO for hyperparameter learning. However, we will discuss how the marginal likelihood is answering a fundamentally different question than "will my trained model provide good generalization?". We consider the discrepancies and their significant practical implications in detail, as well as possible resolutions.

Moreover, it is often thought that Bayesian methods, representing epistemic uncertainty, ought to have more reasonable predictive distributions under covariate shift, since these points will be far from our data manifold. However, we were surprised to find that high quality approximate Bayesian inference often leads to significantly decreased generalization performance. To understand these findings, we investigate fundamentally why Bayesian model averaging can deteriorate predictive performance under distribution and covariate shifts, and provide several remedies based on this understanding.

Bio

Andrew Gordon Wilson is an Associate Professor at the Courant Institute of Mathematical Sciences and Center for Data Science at NYU. Andrew's interests include probabilistic modelling, Gaussian processes, Bayesian statistics, physics inspired machine learning, and model construction and generalization in deep learning. His webpage is here. Andrew is the tutorial chair for NeurIPS 2022, and was the EXPO Chair for ICML 2019, 2020. Andrew has received several awards, including the NSF CAREER Award, the Amazon Research Award, and several best paper, reviewer, and dissertation awards.