what we do
Integrating brain research and artificial intelligence.
Our research seeks to understand how important cognitive behaviors such as learning, remembering, and deciding—which unfold over minutes, hours, days, and even years—are accomplished by the much more rapid, cooperative activity of neurons and synapses in the brain.
The Rajan lab blends brain research and artificial intelligence by developing integrative theories to describe the mechanisms by which cognitive behaviors emerge from underlying neural processes using two primary tools:
Neural network models flexible enough to accommodate sufficient levels of biological detail at the neuronal, synaptic, circuit, and multi-region levels
New and existing mathematical or computational frameworks designed to extract essential mechanistic features in data from well-designed imaging, electrophysiology, and behavioral experiments.
key questions
These theories facilitate critical insights into the learning and execution of cognitive actions, ranging from working memory to high-level phenomena such as reasoning and intuition. Within this broad theme, we are seeking answers to three key questions in neuroscience:
01
How do neurons and circuits in multiple interacting brain regions communicate?
01
This deals with modularity. While many computations in the brain that subserve emotional and cognitive functions require interactions across multiple brain areas, little is known about how such inter-area communication works.
02
How do animals and humans perform a huge range of complex tasks and behaviors?
02
This tackles the brain’s multitasking functionality. Often the same neural circuits are activated in different ways to perform computations on numerous, hard-to-quantify, abstract cognitive variables—processes that will shed light on how the brain categorizes and uses information.
03
How do our brains learn from few examples, solve unstructured problems, and generalize with minimal supervision?
03
This explores the brain's remarkable ability to learn efficiently from limited data. Neural circuits generalize by extracting patterns and principles that apply across various contexts, allowing for flexible problem-solving and adaptation to new situations, but little is known about precisely how this generalization happens.
why does theory matter?
To support these properties—inter-area communication, an impressive task repertoire, and the ability to achieve generalized learning—neural circuits in the contributing brain areas have to be flexible and adaptive to different extents. These circuits are rich and complex in terms of the population dynamics they produce and their underlying circuit and biophysical motifs. As such, the signals collected from these neural circuits through new technologies are also necessarily complex, i.e., multiplexed across different spatial and temporal scales. Principled theory is, therefore, crucial to making sense of these data, guiding new experimental design, and addressing knowledge gaps such as what signals the data contain, over what timescales these signals vary, where signals originate, and how these are similar and different in humans and other animals. Our long-term goal is to perform world-class research leading to integrative theories of brain function underlying consciousness by pioneering and applying new types of mathematical models in tandem with advanced data analysis techniques.
The functions we study are extraordinarily complex
To support inter-area communication, an impressive task repertoire, and the ability to achieve generalized learning, neural circuits have to be flexible and adaptive. These circuits are rich and complex in terms of the population dynamics they produce and their underlying circuit and biophysical motifs. As such, the signals collected from these neural circuits through new technologies are also necessarily complex, i.e., multiplexed across different spatial and temporal scales.
Dense data requires powerful modeling solutions
Our models are designed to use only the biophysical properties of synapses and neurons that form the nervous system. The models we build have the power to extend beyond the details of a single experiment, task, or brain region, revealing unexpected design principles of the real brain.
Where theory can take us
To allow for animals' generalized learning and impressive task repertoires, the brain areas involved in cognition must be flexible and adaptive. The activity patterns and underlying biophysical motifs must be highly complex to support that flexibility. Theory gives us the power to make sense of these data, guide new experimental design, and address knowledge gaps.
Publications
Our publications span studies on recurrent neural networks, reinforcement learning, and modeling tools for the biological brain.
Join us
Join our diverse team that bridges the fields of brain research and artificial intelligence.
%20(1)%20(1).png){kind=logo}