Kanishk Jain
I am a postdoctoral AI researcher at Emory University. My research spans representation learning of complex, high-dimensional data, neural dynamics, and test-time compute, with a focus on building scalable models that reveal structure in data and connect research ideas to working systems.
I am currently a postdoctoral researcher in Tankut Can's group, where I study controllable test-time compute, prompt re-tokenization, and open-weights language model evaluation.
Previously, I completed my PhD in Physics at Emory with Gordon Berman, developing recurrent neural network based pipelines for behavior and gait dynamics.
Research
I am interested in compact representations of complex temporal structure: how to extract interpretable signals from noisy, high-dimensional measurements, and how those signals can reveal individual differences, physiological states, or controllable model behavior.
My recent work focused on prompt re-tokenization in LLMs to generate output diversity, and scalable test-time compute experiments with open-weights language models.
Selected Publications
- Generating output diversity from prompt re-tokenization. Workshop on Scientific Methods for Understanding Deep Learning, ICLR 2026. link
- Discovering individual-specific gait signatures from data-driven models of neuromechanical dynamics. PLOS Computational Biology, 19(10), 2023. doi:10.1371/journal.pcbi.1011556
- Opening the black box of social behavior. Nature Neuroscience, 22, 1947-1948, 2019. doi:10.1038/s41593-019-0547-4
- Anticipating persistent infection. EPL, 121(6), 2018. doi:10.1209/0295-5075/121/60001
- Evolutionary adaptations of biofilms infecting cystic fibrosis lungs promote mechanical toughness by adjusting polysaccharide production. npj Biofilms and Microbiomes, 2016. doi:10.1038/s41522-016-0007-9