The videos from Brain Day 2023 are now available on line at our youtube channel. Hope you enjoy.
Colloquium Series
Here is a link for the speakers for the 2012-2013 year.
- Sept. 11, 2012 - Leonard Maler (University of Ottawa)
- Oct. 23, 2012 - Jan Lauwereyns (Kyushu University, Japan)
- Nov. 6, 2012 - Paul Schrater (University of Minnesota, USA)
- Dec. 18, 2012 - CTN PostDocs (double-header): James Bergstra (Banting Postdoc) & Julien Catanese (FYSSEN Postdoc)
- Jan. 15, 2013 - Tal Yarkoni (University of Colorado at Boulder)
- Feb. 26, 2013 - Veronique Bohbot (McGill)
- Mar. 26, 2013 - Joshua Berke (University of Michigan)
- Apr. 9, 2013 - Brain Day (7th Annual)
- Apr. 23, 2013 - Marc Bellemare (University of Alberta)
Date: Tues., Sept 11, 2012
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Leonard Maler (University of Ottawa)
Title: Multiple Encoding Strategies and the Computational burden of Decoding for Perception
Abstract: Sensory systems of vertebrates typically have to encode many distinct classes of signals emanating from conspecifics (communication), food sources, predators, and environmental features required for navigation (landmarks). Neuronal coding schemes associated with these signal classes include labeled line coding, spike rate coding, spike pattern coding, and population coding. Natural questions that arise are whether there is any connection between the various classes of sensory signals and the different coding types, and what neural dynamics might implement these encoding/decoding strategies. The electrosensory/motor system offers is a useful preparation to study such problems because it is relatively simple to define and mimic the natural signals associated with communication, prey capture and navigation. Behavioral studies have revealed that the fish responds to these signal classes and must therefore have the neuronal machinery for their encoding, perceptual classification, and decoding into motor activity. Presenting the fish with mimics of natural signals while recording spiking responses have allowed us to trace encoding transformations from electroreceptors to midbrain neurons. We find that many disparate encoding strategies are differentially used to encode signals associated with communication, prey detection and landmark recognition. Labeled line coding is generally not present at early sensory processing stages, but emerges in the midbrain as a result of code sparsening. At early processing stages various combinations of spike patterning, population coding and encoding via complex temporal sequences of spiking responses across a population are required to connect the electroreceptor responses and the final behavioral outcome of sensory stimulation.
Date: Tues., Oct. 23, 2012
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Jan Lauwereyns (Kyushu University, Japan)
Date: Tues., Nov. 6, 2012
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Paul Schrater (University of Minnesota, USA)
Date: Fri., Dec. 18, 2012
Location: PAS 2464
Time: 3:30 p.m.
Speaker: CTN PostDocs (double-header): James Bergstra (Banting Postdoc) & Julien Catanese (FYSSEN Postdoc)
Date: Tues., Jan. 15, 2013
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Tal Yarkoni (University of Colorado at Boulder)
Date: Tues., Feb. 26, 2013
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Veronique Bohbot (McGill)
Date: Tues., Mar. 26, 2013
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Josh Berke (University of Michigan)
Date: Tues., April 23, 2013
Location: PAS 2464
Time: 3:30 p.m.
Speaker: Marc Bellemare (University of Alberta)
Title: Pixels and Priors: Learning a Generative Model of Atari 2600 Games
The model-based approach to reinforcement is appealing as it allows the agent to reason about unexplored parts of its environment. However, most existing model learning techniques impose harsh restrictions on the model structure or simply do not scale well to large domains. In this presentation, I will present a new algorithm for learning generative models of arbitrary Atari 2600 games. Using statistical data compression methods, this algorithm achieves a per-step cost linear in the number of pixels, making it an interesting alternative to other model learning methods. One of the novel components of this algorithm is the quad-tree factorization, which uses Bayesian model averaging to efficiently learn a variable-resolution factorization of the Atari 2600 observation space. I will provide redundancy bounds that guarantee the statistical efficiency and soundness of the approach and some recent empirical results on a variety of Atari 2600 games.
