Research map / Architectures and efficiency / Quantization
The Synthesis of XNOR Recurrent Neural Networks with Stochastic Logic
Arash Ardakani, Zhengyun Ji, Amir Ardakani, Warren Gross
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 |
|---|---|---|---|
| binarize state computations with stochastic computing (XNOR-LSTM) instead of full-precision state (ablation) | ELSTM · PTB character-level LM | BPC: 1.47 -> 1.52 | Hurt |
| binarize state computations with stochastic computing instead of full-precision state (ablation) | ELSTM · PTB word-level LM | perplexity: 93.5 -> 95.5 | Hurt |
| binarize state computations with stochastic computing instead of full-precision state (ablation) | ELSTM · CNN QA | error rate: 40.4 -> 43.8 | Hurt |
| binarize state computations with stochastic computing instead of full-precision state (ablation) | ELSTM · PTB character-level LM | cost (XNOR count): 604,000 -> 7,000 | Improved |
| use XNOR-LSTM instead of HitNet | quantized LSTM · PTB word-level LM | perplexity: 110.3 -> 95.5 | Improved |
| use XNOR-LSTM instead of LAB | quantized LSTM · PTB character-level LM | BPC: 1.56 -> 1.52 | Improved |
| replace IStanh with weighted-state FSM function WIStanh (ablation) | FSM approximation of tanh and sigmoid · PTB neuron outputs | approximation of tanh/sigmoid | Improved |
| use XNOR-LSTM instead of non-stochastic binarized LSTM on FPGA | hardware implementation · Virtex-7, 300 neurons | area and throughput: 66K vs 1.1M slices; 3.2 vs 1.8 TOPS | 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.
- HitNet — HitNet: Hybrid Ternary Recurrent Neural Network (NeurIPS 2018) · better on 1 of 1 shared result
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
- Learning Recurrent Binary/Ternary Weights (ICLR 2019 · similarity 0.86)
- xLSTM: Extended Long Short-Term Memory (NeurIPS 2024 · similarity 0.85)
- Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity (ICML 2025 · similarity 0.85)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · similarity 0.84)
- Analog Foundation Models (NeurIPS 2025 · similarity 0.84)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.84)
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.
