Aravind Srinivas

Papers

Total Citations

251

H-Index

About

Aravind Srinivas is an AI researcher whose work spans generative modeling, reinforcement learning, and visuomotor control, with particular focus on bridging the gap between powerful deep learning architectures and sequential decision-making problems. He is perhaps best known for **VideoGPT** (2021), a landmark contribution that demonstrated how VQ-VAE combined with transformer architectures could scale likelihood-based generative modeling to natural videos — work that has garnered over 140 citations and helped lay conceptual groundwork for the video generation field. His earlier research on **Universal Planning Networks** (2018, 92 citations) tackled one of reinforcement learning's fundamental challenges: learning abstract, goal-directed representations capable of supporting planning and generalization in complex visuomotor tasks. He has also explored how modern deep learning architectural innovations, such as dense connectivity, can improve reinforcement learning agents through his work on D2RL. Across his research portfolio, Srinivas consistently demonstrates an ability to translate ideas from computer vision and natural language processing into richer, more capable sequential models — making him a compelling voice at the intersection of generative AI and autonomous systems. He later co-founded Perplexity AI, applying his research background to real-world AI products.