Assistant Professor - Department of Neuroscience Ph.D., University of Groningen, The Netherlands, 2001 One Baylor Plaza Baylor College of Medicine Houston TX, 77030 Smith Medical Research Bldg Room: T115D Telephone: 713-798-8407 Email: wjma@cpu.bcm.edu Website: neuro.bcm.edu/malab

Sensory information received by the brain is typically uncertain (for instance because of poor signal quality or ambiguity in the world), yet it must constantly be manipulated to generate accurate, task-relevant behavior. In our laboratory, we investigate how the brain represents and processes uncertain information. This is critical for approaching a key problem in systems neuroscience, that of understanding the relationship between perceptual behavior and neural activity.

There is good evidence from human studies that the brain weighs and integrates pieces of information in a way that takes into account their uncertainty. (In the jargon of the field, this is called “Bayes-optimal” performance.) We study such behaviors using computational modeling and psychophysics in a variety of domains, including multisensory perception, decision-making, and visual search. We are particularly interested in pushing the boundaries of Bayesian optimality by considering complex stimuli (like auditory-visual speech) and complex computations (like those involved in visual search, change detection, or causal inference).

At the neural level, a leading idea is that neural populations maintain representations of uncertainty (or even of entire probability distributions) and use these to perform optimal computation. We use theories of neural coding and large-scale simulations of biologically realistic neural networks to study how optimal computations can be implemented in the brain. For example, we have shown that optimal cue integration can be achieved through relatively simple neural operations, provided that trial-to-trial neural variability is of so-called Poisson-like form. This general approach has the potential to make physiologically testable predictions in many areas of perception and cognition.

top

Ma WJ, Huang W (2009), No capacity limit in attentional tracking: evidence for probabilistic inference under a resource constraint. Journal of Vision 9 (11): 3, 1-30.

Ma WJ, Zhou X, Ross LA, Foxe JJ, Parra LC (2009), Lip-reading aids word recognition most in moderate noise: a Bayesian explanation using high-dimensional feature space. PLoS ONE 4 (3): e4638.

Beck JM, Ma WJ, Kiani R, Hanks TD, Churchland AK, Roitman JD, Shadlen MN, Latham, PE, and Pouget A (2008), Bayesian decision making with probabilistic population codes. Neuron 60 (6), 1142-5.

Ma WJ, Pouget A (2008), Linking neurons to behavior in multisensory perception: a computational review. Brain Research 1242, 4-12.

Ma WJ, Beck JM, Pouget A (2008), Spiking networks for Bayesian inference and choice. Current Opinion in Neurobiology 18, 217-22.

Beierholm U, Kording K, Shams L, Ma WJ (2007), Comparing Bayesian models for multisensory cue combination without mandatory integration. Advances in Neural Information Processing Systems.

Beck JM, Ma WJ, Latham PE, Pouget A (2007). Probabilistic population codes and the exponential family of distributions. Progress in Brain Research 165, 509-19.

Kording K, Beierholm U, Ma WJ, Quartz S, Tenenbaum JB, Shams L (2007), Causal inference in multisensory perception. PLoS ONE 2 (9), e943.

Ma WJ, Beck JM, Latham PE, Pouget A (2006). Bayesian inference with probabilistic population codes. Nature Neuroscience 9 (11), 1432-8.

Ma WJ, Hamker F, Koch C (2006). Neural Mechanisms Underlying Temporal Aspects of Conscious Visual Perception. in The First Half Second: The Microgenesis and Temporal Dynamics of Unconscious and Conscious Visual Processes, Ögmen H, Breitmeyer BG eds., 275-94, MIT Press.

top

Optimal integration of two sensory cues using probabilistic population codes. The cues evoke activity in input populations, denoted by vectors r1 and r2, and indicated by green and blue dots. Neurons are ordered by their preferred stimulus. A simple linear combination of input population patterns of activity, r3 = W1r1 + W2r2 (shown in red dots), guarantees optimal cue integration, if neural variability is so-called Poisson-like. This includes independent Poisson variability, but also allows for correlated variability. The dialogue boxes show the probability distributions over the stimulus encoded in the populations on a single trial. Optimal cue integration is characterized by a multiplication of probability distributions over the stimulus, p(s|r3) µ p(s|r1)p(s|r2). W1 and W2 are synaptic weight matrices which depend on the statistics of the input populations, but do not have to be adjusted over trials. This computation can also be implemented using biologically realistic neurons. For more information, see Ma et al., Nature Neuroscience 2006.

top