We use principles of efficient coding (normative models – see below) to understand sensory processing in the brain. We specialize in understanding early visual development through an “innate learning” paradigm – the training of the early visual system on endogenously generated neural patterns.
- April 30, 2014: See Gordon Kratz’s talk on his funded summer research effort “Binocular depth perception scoring for visual development models” at the CS research presentations
- April 14, 2014: Math Colloquium talk by Dr. Albert: “How to Derive a Brain: efficient coding in sensory neuroscience”
- April 10, 2014: CIT grant accepted for Loyola’s Summer Research Experience in Neuroscience – proposed by Toby Dye and Robert Morrison for Summer 2015
- April 7, 2014: Congrats to Gordon Kratz for receiving the Provost Fellowship for his project “Augmented depth perception as a behaviorally-relevant metric for the innate learning paradigm”
- Nov 12, 2013: Society for Neuroscience 2013 poster presentation: Mobility in PD improved through dance.
- Nov 6, 2013: Biology seminar: “Innate learning models of LGN/V1 spontaneous activity” (11:30am, location: TBD)
- Sep 3, 2013: Neuroscience seminar: “Innate learning in early visual development” (4pm, Cuneo 109)
- Dec 6, 2012: Oculomotor adaptation paper published in Frontiers in Computational Neuroscience. [pdf]
- Nov 15, 2011: Presentation accepted for Neuroscience 2011 Using the statistics of binocular images to model spontaneous activity in the developing visual system [abstract]
Normative modeling, explained:
One can attempt to understand the visual system by directly measuring neural responses to stimuli, and characterizing the responses of each cell type. A vast amount of data has been and will continue to be collected in this way using fMRI, electrophysiology, EEG, etc. However, it is of prime importance to make sense of this data in a way to make it useful in other domains (e.g. AI, machine learning). A normative approach in this case would be to understand the complicated response properties as an efficient encoding of visual experience. Instead of using complicated mathematical constructs to model a neuron’s response, one can use a simple algorithm to generate the complicated response properties we see. Many modeling approaches answer the detailed questions of “what” or “how” for neural responses, but normative models also help to answer “why”. This lab uses these high-level principles to explain the varied properties of sensory neural response.