Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University

Reasoning has become a defining capability of large language models, yet the learning mechanisms behind its emergence and generalization remain poorly understood. Yu Huang examines how chain-of-thought training on short problems enables length generalization, and how self-improvement can extend that ability further. The talk then analyses how the reasoning frontier evolves under reinforcement learning with verifiable rewards, revealing an implicit curriculum that emerges from mixed-difficulty training and showing how the spectrum of problem difficulty shapes learning dynamics. The talk is one of the Kempner Institute Research Fellow candidate presentations.

Yatin Dandi traces a line of work on how neural networks learn features, from hidden directions in two-layer models to the intermediate representations that let deeper networks learn complex targets with far fewer samples. A unifying theme is that networks learn features through low-degree correlations in the data. This insight led to Neural Low-Degree Filtering, a spectral algorithm that describes early layer-wise learning on real data, and to its extension to the joint training of layers. The talk closes with approaches to understanding the full dynamics of hierarchical learning across layers, drawing on a rigorous version of the dynamical cavity method. The talk is one of the Kempner Institute Research Fellow candidate presentations.

Most of artificial intelligence, and most models of the brain, treat a neuron as a simple unit that sums its inputs. Real cortical neurons, however, carry elaborate dendritic trees with nonlinear ion channels and thousands of synapses, which points to substantial computation inside a single cell. Ido Aizenbud trains biophysically detailed, realistic neuron models with gradient descent and shows that one cortical neuron can solve sensory tasks such as visual and auditory classification, as well as abstract nonlinear parity tasks that lie far beyond the reach of point neurons. The talk then outlines how this approach scales to networks of detailed neurons, treating dendrites as architectural primitives both for understanding brain circuits and for building more powerful, energy-efficient AI. The talk is one of the Kempner Institute Research Fellow candidate presentations.