Exploiting neuro-inspired dynamic sparsity for energy-efficient intelligent perception
in Nature Communications, 2025
Compute only what changes.
Sparse neural computing
Neural activity changes over time. We study how to take advantage of those changes in recurrent networks, training methods and hardware architectures. Our work spans DeltaRNN, EdgeDRNN and Spartus, and broader perspectives on dynamic sparsity for efficient intelligent perception.
Selected work.
All publicationsSpartus: A 9.4 TOp/s FPGA-Based LSTM Accelerator Exploiting Spatio-Temporal Sparsity
in IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2024
DeltaRNN: A Power-efficient Recurrent Neural Network Accelerator
In Proceedings of the 2018 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA '18). Association for Computing Machinery, New York, NY, USA, 21–30.
Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training
accepted to the 38th Annual AAAI Conference on Artificial Intelligence (AAAI), 2024
DeltaLLM: A Training-Free Framework Exploiting Temporal Sparsity for Efficient Edge LLM Inference
preprint, 2025.
To Spike or Not To Spike: A Digital Hardware Perspective on Deep Learning Acceleration
in IEEE Journal on Emerging and Selected Topics in Circuits and Systems (JETCAS), 2023 (JETCAS Spotlight Article 2023)