Upcoming Thesis Defenses

July 5, 2022
10:00 a.m., Meliora 269 & Zoom meeting

Thesis Defense: Joseph German, Brain & Cognitive Sciences (Advisors: Robbie Jacobs and Len Schubert)

Explaining the Flexible Use of Conceptual Knowledge in Human Cognition and Runtime Learning

We introduce the “runtime learning hypothesis”, which states that people quickly learn to perform unfamiliar tasks as the tasks arise by using task-relevant instances of concepts stored in memory during mental training. To make learning rapid, the hypothesis claims that only a few class instances are used, but these instances are especially valuable for training. The paper motivates the hypothesis by describing related ideas from the cognitive science and machine learning literatures. Using computer simulation, we show that deep neural networks (DNNs) can learn effectively from small, curated training sets, and that valuable training items tend to lie toward the centers of data item clusters in an abstract feature space. In a series of three behavioral experiments, we show that people can also learn effectively from small, curated training sets. Critically, we find that participant reaction times and fitted drift rates are best accounted for by the confidences of DNNs trained on small datasets of highly valuable items.

In order to extend the runtime learning hypothesis, we then explore the area of few-shot learning and relate it to the human capacity to “learn to learn”. We perform a proof-of-concept examination of the ability of runtime learning to make use of various few-shot learning algorithms to account for human behavioral data. We also speculate on how additional work in few-shot learning, in particular more challenging datasets and training algorithms that encourage feature adaptation, may improve this explanatory ability.

In light of the importance of similarity judgments in few-shot learning, we next describe and analyze the performance of metric learning systems, including deep neural networks (DNNs), on a new dataset of human similarity judgments of “Fribbles”, naturalistic, part-based objects. Metrics trained using pixel-based or DNN-based representations fail to explain our experimental data, but a metric trained with a viewpoint-invariant, part-based representation produces a good fit. We also find that although neural networks can learn to extract the part-based representation—and therefore should be capable of learning to model our data—networks trained with a “triplet loss” function based on similarity judgments do not perform well. We analyze this failure, providing a mathematical description of the relationship between the metric learning objective function and the triplet loss function. The comparatively poor performance of neural networks appears to be due to the nonconvexity of the optimization problem in network weight space. We discuss the implications for neural network research as a whole.


Past Thesis Defenses

2021

December 6, 2021

Shanna Coop, Brain & Cognitive Sciences, University of Rochester

Neural Mechanisms of Presaccadic Attention and Foveal Prediction


October 12, 2021

Linghao Xu, Brain and Cognitive Sciences, University of Rochester

Models of Approximate Inference in Vision (MA thesis defense)


September 14, 2021

Ankani Chattoraj, Brain and Cognitive Sciences, University of Rochester

Models of Approximate Inference in Vision


June 18, 2021

Annette French, Brain and Cognitive Sciences & MSTP, University of Rochester

Perceptual Processes Underlying Depth Judgement of Moving Object During Self-Motion


August 5, 2021

Adam Danz, Brain and Cognitive Sciences, University of Rochester

Predictive Steering Control and Neuronal Representation of Heading in Macaques


2020

March 30, 2020

Zheng Liu, Biomedical Engineering, University of Rochester

Neural Activity in Multiple Cortical Areas during Brain-Computer Interface Control


May 12, 2020

Shirlene Wade, Brain and Cognitive Sciences, University of Rochester

Situational Determinants of Curiosity: An experimental approach


May 26, 2020

Andres Guevara Torres, Institute of Optics, University of Rochester

Imaging Translucent Retinal Neurons and Vascular Cells by Optimizing Phase Contrast in the Living Mouse Eye


June 9, 2020

Sunwoo Kwon, Brain and Cognitive Sciences, University of Rochester

Understanding How Pre-Saccadic Attention Influences Predictive Oculomotor Behavior and the Underlying Neural Circuitry


June 23, 2020

Colleen Schneider, Brain and Cognitive Sciences, University of Rochester

The post-stroke brain: A comprehensive, longitudinal assessment of vascular, neural, and perceptual changes


July 21, 2020

Chris Bates, Brain and Cognitive Sciences, University of Rochester

Efficient Data Compression in Human Perception


July 31, 2020

Nicole Peltier, Brain and Cognitive Sciences, University of Rochester

Neural Basis of Object Motion Perception During Self-Motion


August 25, 2020

Richard Lange, Brain and Cognitive Sciences, University of Rochester

Signatures of Approximate Bayesian Inference in Early Visual Perception


2019

January 30, 2019

Matthew Overlan, BCS Graduate Student, University of Rochester

Probabilistic Program Induction as a Model of Human Concept Learning


February 6, 2019

Mengchen Xu, CVS Graduate Student, University of Rochester

Investigation of Corneal Biomechanical and Optical Behaviors by Developing Individualized Finite Element Model


April 5, 2019

Sarah Walters, Graduate Student, Institute of Optics, University of Rochester

Two-Photon Excited Fluorescence Adaptive Optics Ophthalmoscopy of Retinal Function


August 23, 2019

Elizabeth Shay, Graduate Student, University of Rochester

Neural Signatures of Compositionality in the Human Brain


December 9, 2019

Elizabeth Saionz, Graduate Student, Translational Biomedical Science with the Center for Visual Science, University of Rochester

Time is Vision: Properties of Vision Early after Occipital Stroke and Capacity for Recovery


2018

June 4, 2018

Kim Schauder, Graduate Student, CSP, University of Rochester

Initial Eye Gaze to Faces and their Functional Consequence on Face Identification Abilities in Autism Spectrum Disorder


June 20, 2018

Quanjing Chen, Graduate Student, BCS, University of Rochester

The Representation of Tool Knowledge in the Human Brain


August 20, 2018

Michael Melnick, Graduate Student, University of Rochester

Neural Efficiency and Processing: Understanding Visual Perceptual Processing using Transcranial Electrical Stimulation and fMRI in Cortically Blind Humans


December 13, 2018

Alexander Kotelsky, PhD candidate, Department of Biomedical Engineering, University of Rochester

Elucidating the Factors That Govern Vulnerability of In situ Articular Chondrocytes to Impact and Sub-impact Mechanical Loading


2017

April 3, 2017

Goker Erdogan, BCS Graduate Student, University of Rochester

Shape Perception as Bayesian Inference of Modality-Independent Part-Based 3D Object-Centered Shape Representations


July 26, 2017

Frank Garcea, Graduate Student, BCS, University of Rochester

The Organization of Manipulable Object Concepts in the Human Brain


September 6, 2017

Woon Ju Park, Graduate Student, University of Rochester

A Mechanistic Understanding of Atypical Visual Processing in Autism Spectrum Disorder