Research map / Architectures and efficiency / Spiking networks
SmoothSpike: Spiking Transformer with Learnable Hadamard Transformation
Zijian Zhou, Wenjie Wei, Yu Liang, Jialin Li, Ammar Belatreche, Honglin Cao, Shuai Wang, Malu Zhang, Yang Yang, Haizhou Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SmoothSpike to Spikingformer | Spikingformer · GLUE dev | average score: 66.8 -> 75.0 | Improved |
| add SmoothSpike to SpikeLM | SpikeLM · GLUE dev | average score: 75.7 -> 77.5 | Improved |
| add SmoothSpike to SpikeLM | SpikeLM · GLUE RTE | score: 65.3 -> 58.1 | Hurt |
| add SmoothSpike to Spikingformer | Spikingformer · GLUE STS-B | score: 54.5 -> 79.3 | Improved |
| add SmoothSpike to Spikingformer | Spikingformer · GLUE CoLA | score: 24.4 -> 41.2 | Improved |
| add SmoothSpike to Spikingformer | Spikingformer · GLUE energy | energy (mJ): 6.76 -> 9.45 | Hurt |
| add learnable Hadamard H1 to Spikingformer (ablation) | Spikingformer baseline · GLUE dev ablation | average score: 64.00 -> 65.88 | Improved |
| add learnable H2 after H1 (ablation) | Spikingformer + learnable H1 · GLUE dev ablation | average score: 65.88 -> 67.64 | Improved |
| add learnable H3 after H1 and H2 (ablation) | Spikingformer + learnable H1,H2 · GLUE dev ablation | average score: 67.64 -> 67.88 | Improved |
| add fixed randomized Hadamard H1 instead of nothing (ablation) | Spikingformer baseline · GLUE dev ablation | average score: 64.00 -> 63.74 | No clear effect |
| make H1 learnable instead of fixed (ablation) | Spikingformer + H1 · GLUE dev ablation | average score: 63.74 -> 65.88 | Improved |
| apply H3 learnable Hadamard transform (ablation) | Spikingformer · saturated neuron rate (Layer 4) | saturation rate (%): 3.79 -> 0.60 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Spikingformer — Spikingformer: A Key Foundation Model for Spiking Neural Networks (AAAI 2026) · better on 9 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SpikeLLM: Scaling up Spiking Neural Network to Large Language Models via Saliency-based Spiking (ICLR 2025 · similarity 0.88)
- SpikeLM: Towards General Spike-Driven Language Modeling via Elastic Bi-Spiking Mechanisms (ICML 2024 · similarity 0.88)
- Rethinking Spiking Neural Networks from an Ensemble Learning Perspective (ICLR 2025 · similarity 0.88)
- Towards Efficient Spiking Transformer: a Token Sparsification Framework for Training and Inference Acceleration (ICML 2024 · similarity 0.88)
- STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking (NeurIPS 2025 · similarity 0.88)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
