CLASS NOTES, SLIDES, READING MATERIAL, AND VIDEO
01/23/2024 Gabriel Kreiman Introduction to biological and artificial intelligence Are you ready to create suprahuman intelligence? Do you want to understand how intelligent computations emerge from the neuronal orchestra in the brain? How can we leverage millions of years of evolution to transform engineering systems that perform intelligent computations? How concerned should we be about the singularity? How is information represented and transformed in biological circuits? How can machines learn?
Reading Video (ID required, 2020) 01/30/2024 Richard Born Biological and artificial vision We understand the biological and computational underpinnings of our prodigious ability to see better than any other aspect of our brain's function. Beginning with the seminal work of Hubel and Wiesel, we have discovered many of the biological principles that operate in the primate visual system. And computational models inspired by these principles have enjoyed tremendous success in object recognition and other visual capacities, sometimes outperforming humans. Yet these models also fail in spectacular and occasionally embarrassing ways, leading to the overwhelming question: What are they missing? We will explore the history of biological and machine vision, learn how they productively interact in present-day vision science and discuss whether Mother Nature might still have a few tricks up her sleeve that could spur the development of improved machine vision systems in the future.
Slides Reading No video available 02/06/2024 Jan Drugowitsch
The Bayesian brain: ideal observer models for perceptual decisions
The aim of both biological and artificial intelligence is to find efficient solutions to hard problems. Ideal observer models use this parallel to formulate theories of how our nervous system achieves certain tasks based on how they have been approached by artificial intelligence research. A basic tenant of ideal observer models is that the world we inhabit is noisy and ambiguous. The ideal way to deal with the arising uncertainty is by Bayesian decision theory. I will introduce Bayesian decision theory and how it has been used to inquire about the cognitive and neural processes that underlie humans and animal behavior. Specific focus will be put on the study of perceptual decisions and the relation between the speed and accuracy of such decisions.
Slides Reading Video (ID required, 2020) 02/13/2024 Kanaka Rajan Unlocking Brain Dynamics: The Power of Recurrent Neural Networks in Neuroscience
Dr. Rajan will discuss the basic design elements, or "building blocks," of neural network models, highlighting the role of recurrent connections, linear and non-linear activity, and the types of time-varying activity ("dynamics") produced by RNNs, including input-driven versus spontaneous dynamics. The main goal of this lecture is to explore both the tractability and computational power of RNN models, to appreciate why they have become a crucial part of the neuroscientific arsenal, and to understand the insights that can be gained by "training RNNs to do something" in a manner consistent with experimental data from the biological brain. We will also review how RNNs have been applied in neuroscience to leverage existing experimental data, infer mechanisms inaccessible from measurements alone, and make predictions that guide experimental design.
Slides Reading No video available 02/20/2024 Haim Sompolinsky The geometry of concept learning This class will present a theory of manifold separation in high-dimensional spaces, few-shot learning in visual representations, and aspects of visual reasoning.
Slides Reading No video available 02/27/2024 Sam Gershman Using video games to reverse engineer human intelligence Video games have become an attractive testbed for evaluating AI systems, by capturing some aspects of real-world complexity (rich visual stimuli and non-trivial decision policies) while abstracting away from other sources of complexity (e.g., sensory transduction and motor planning). Some AI researchers have reported human-level performance of their systems, but we still have very little insight into how humans actually learn to play video games. This talk will present new data on human video game learning indicating that humans learn very differently from most current AI systems, particularly those based on deep learning. Humans can induce object-oriented, relational models from a small amount of experience, which allow them to learn quickly, explore intelligently, plan efficiently, and generalize flexibly. These aspects of human-like learning can be captured by a model that learns through a form of program induction.
Slides Reading No video available 03/05/2023 Tomer Ullman
The development of intuitive physics and intuitive psychology
The central metaphor of cognitive science is that of the mind as a computer, but what sort of program is the mind running, and how does it construct this program? From an evolutionary perspective it would make sense to build in certain primitives and functions to allow the mind to get an 'early start' on understanding the world. These primitives would most helpfully be those that are generalizable across many scenarios, such as an understanding of people and things, agents and objects, psychology and physics. And indeed, we can see early evidence for an understanding of physics and psychology even in young children. I will briefly go over the evidence for an early understanding of physics and psychology, what representations could account for that understanding, and how they may develop over time into adult representations.
Slides Reading Video (ID required, 2020) 03/12/2024 No classes Spring Break 03/19/2024 Andrei Barbu Language in brains and machines Slides Reading No video available 03/26/2024 Thomas Serre State of the art in computer vision
Slides Reading No video available 04/02/2024 Cengiz Pehlevan
Why is it that biological and artificial neural networks do not overfit?
Slides Reading No video available 04/09/2024 Gabriel Kreiman Ethics, opportunities and the future of AI (Discussion/Debate)
AMA session to discuss what we have been seeing in class, ethical regulations, interpretability, explainability and future of AI research.
Slides Reading No video available 04/16/2024 Will Xiao Robustness in biological and artificial vision | Discussion on neurosymbolic AI As a fellow student of biological and artificial intelligence, I will share as working examples some of my past, ongoing, and emerging research interests in the two areas. After synopsizing some past work, the first hour will scrutinize the topic of robustness to small-pixel-norm, adversarial-like image noise. This topic gains interest from the dissonance between two current views: that (naïve) deep nets are highly sensitive to adversarial images, whereas biological vision seems robust, and that deep nets provide the current best models of biological visual representations. I will discuss continuing work by others and myself to quantify biological vision robustness and relate it to the (dis)similarities between biological and artificial vision. The second hour will survey research directions loosely grouped under neurosymbolic AI. I will discuss select studies that show promise in synergizing connectionist and symbolic AI ideas toward high-performance yet sample-efficient algorithms that can learn, generalize, and ‘think like people do.’ I will also discuss neuroscience approaches to studying symbolic/abstract thinking in the brain.
Slides Reading No video available 04/23/2024 Student presentations Final project presentations Reading 04/30/2024 Student presentations Final project presentations Reading
