Arindam Mitra
Papers
1
Total Citations
4
H-Index
1
About
Arindam Mitra investigates the intersection of neural networks and formal reasoning, with a particular focus on whether large language models can emulate logical deduction. His most-cited work, "Can Transformers Reason About Effects of Actions?" (2020, 4 citations), challenges the field by probing the limits of transformer architectures in handling rule-based inference. Mitra’s research demonstrates that while transformers can reason with facts and rules expressed as natural language conjunctions of conditions implying conclusions, their capabilities are constrained to narrow, well-defined settings. This foundational contribution has sparked critical discussions about the true nature of machine reasoning versus pattern matching. By systematically testing transformers on action-effect reasoning tasks, Mitra has helped clarify both the promise and the pitfalls of using neural models for knowledge-driven inference. His work is essential reading for researchers exploring neuro-symbolic AI, causal reasoning, and the integration of structured knowledge into deep learning systems.
Research Focus
Key Achievements
Top Papers
- 1Can Transformers Reason About Effects of Actions?4 citations · 2020