Fellows seminar - Representation of information in the brain

By Arvind Kumar, 2022 Fellow

How internal and external information is represented in the brain is one of the fundamental questions in neuroscience. Experiments have shown that neurons are sensitive to a small set of features, e.g. a neuron in the visual cortex may respond to color and orientation angle. Moreover, neurons respond to only a small range of the variables they prefer. For instance, some neurons in the primary visual cortex respond to moving bars oriented at specific angles. This has led to the notion of ‘tuning curve’.

Experiments have revealed a wide range of tuning curve shapes for a variety of stimuli. In my talk, I will discuss how the shape of the tuning curves impose constraints on fidelity information representation in the brain. First, I will develop a heuristic about the optimal shapes of tuning curves given noise and correlations. This perspective already gives us a putative explanation of why different brain regions may prefer a certain type of tuning curve shapes. 

Next, I will compare the performance of two prominent classes of tuning curves, examining not only their accuracy but also the time required to achieve a certain level of accuracy. This analysis reveals a trade-off between speed and accuracy - with some tuning curves shapes, neurons quickly reach a minimum accuracy level but may lack precision, whereas neurons with more accurate tuning curves may take longer to reach their minimum accuracy. Tuning curve shapes are primarily determined by neuron connectivity, with learning influencing these shapes quantitatively. Consequently, it appears that the inherent speed-accuracy trade-offs crucial for natural behavior are wired into the brain's connectivity.

• More information on Arvind Kumar and his USIAS project: Functional implications of bidirectional connectivity between cerebellum and basal ganglia