Anish Mudide

I'm a freshman at MIT studying Math and Computer Science. I research how large language models work and how they can be used to benefit humanity.

I am fortunate to collaborate with Max Tegmark and Bonnie Berger.

Email / Scholar / Twitter / GitHub

Publications

	DafnyBench: A Benchmark for Formal Software Verification Chloe Loughridge, Qinyi Sun, Seth Ahrenbach, [...], Anish Mudide et al. Under review arXiv / code The largest benchmark for AI-assisted formal software verification.
	Causal gene regulatory analysis with RNA velocity reveals an interplay between slow and fast transcription factors Rohit Singh, Alexander Wu, Anish Mudide, Bonnie Berger Cell Systems*, 2024 project page / video / arXiv / code We leverage RNA velocity to build a causal model of gene regulation.
	Opening the AI black box: program synthesis via mechanistic interpretability Eric J. Michaud, Isaac Liao, Vedang Lad, Ziming Liu, Anish Mudide et al. Under review paper / code Distilling algorithms learned by neural networks into Python code.
	The contribution of historical processes to contemporary extinction risk in placental mammals Aryn P. Wilder, Megan A. Supple, Ayshwarya Subramanian, Anish Mudide et al. Science, 2023 paper / code We train the first models capable of predicting conservation status from solely genomic data.
	Generation of LexA enhancer-trap lines in Drosophila by an international scholastic network Ella S. Kim, Arjun Rajan, Kathleen Chang, [...], Anish Mudide et al. G3: Genes, Genomes, Genetics, 2023 paper / arXiv We generate 301 novel LexA enhancer trap insertions and study their tissue expression patterns.
	SARS-CoV-2 drug discovery based on intrinsically disordered regions Anish Mudide, Gil Alterovitz PSB, 2021 paper / code We use molecular docking and train message-passing neural networks to target the disordered proteome of SARS-CoV-2.
	How much does the (social) environment matter? Using artificial intelligence to predict COVID-19 outcomes with socio-demographic data Christos Makridis, Anish Mudide, Gil Alterovitz PSB, 2021 (Oral Presentation) paper / code We find that socio-demographic and social capital characteristics are crucial determinants of public health outcomes.

Blog Posts

Efficient Dictionary Learning with Switch Sparse Autoencoders
Anish Mudide, Josh Engels, Christian Schroeder de Witt

We introduce the Switch Sparse Autoencoder, a new architecture that leverages Mixture-of-Experts to greatly improve the efficiency of dictionary learning.

Replicated from Jon Barron.