ACE: Adaptive Compute-Efficient State Space Models
Dynamic token routing for Mamba models that adapts computation to token importance.
Read research overviewAshiqur Rasul / Research
Research
Input-aware algorithms and efficient hardware, from state space models to in-memory computing. Explore the motivation, approach, and findings behind my recent papers.
CASPER: Adaptive Retrieval with In-DRAM Processing
Confidence-guided dimension adaptivity and in-DRAM similarity search for efficient information retrieval.
Read research overviewAdaptive Testing of Compute-in-Memory GANs
Backpropagation-guided test compaction for faster testing of analog compute-in-memory systems.
Read research overviewEnergy-Efficient Domain-Specific AI Models and Agents
A cross-layer perspective on domain-specific intelligence, brain-inspired computing, and efficient hardware.
Read research overviewDynamic 3D CNN Pruning for Efficient Edge Inference
Input-dependent frame, channel, and feature adaptation to reduce the cost of video inference.
Read research overview