Luke Hewitt, Tenenbaum Lab
In this talk I will describe a tension that runs through cognitive science and AI: the need for representation spaces which are expressive enough to capture the structure of a domain, while ensuring that searching the space remains tractable. I will give examples of how this tension has played out, including in my own work on learning probabilistic programs and deep generative models of handwritten characters. The difficulty of search often arises from the blind, amnesic nature of many general-purpose search algorithms, which stands in stark contrast to the directed way in which humans generate and test hypotheses. Using machine learning to improve search algorithms offers a promising way forward, allowing us to search richer spaces and explore more expressive models.
Maddie Cusimano, McDermott Lab
We often experience sound as a mixture of multiple simultaneous sources. However, this structure is not evident in the sensory data alone, as there are an infinite number of causal explanations for any acoustic signal. This talk will describe a hierarchical Bayesian model of how the auditory system constructs explanations of auditory scenes, focusing on classic perceptual demonstrations of auditory grouping. Despite this emphasis, the model can also make reasonable inferences of source structure from recorded audio. Given the model's ability to generalize, this work provides a foundation for studying scene analysis with natural sounds.
UPCOMING COG LUNCH TALKS:
10/10/17 - Sarah Schwettman (Kanwisher Lab) and Dae Houlihan (Saxe Lab)
10/17/17 - Meilin Zhan (Levy Lab), Morteza Sarafyazd (Jazayeri Lab)
10/24/17 - Maddie Pelz (Schulz Lab), Matthias Hofer (Levy Lab), and Andres Campero (Tenenbaum Lab)
10/31/17 - Mika Braginksy (Levy Lab), Jenelle Feather (McDermott Lab), and Andrew Francl (McDermott Lab)
11/07/17 - Richard McWalter, Ph.D. (McDermott Lab)
11/14/17 - Kevin Ellis (Tenenbaum Lab)
11/21/17 - Dian Yu (Rosenholtz Lab)
11/28/17 - Yang Wu (Schulz Lab)
12/05/17 - Melissa Kline, Ph.D.
12/12/17 - Rachel Magid (Schulz Lab)