Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Standard PEFT methods struggle in multi-task fine-tuning settings due to task interference and representational deficiency.limitation 85%
• Different tasks exhibit distinct frequency energy distributions in LLMs.result 90%
• LLM layers display heterogeneous frequency sensitivities.result 90%
• FourierMoE integrates MoE architecture with inverse discrete Fourier transform for frequency-aware adaptation.method 95%
• FourierMoE employs a frequency-adaptive router to dispatch tokens to experts specialized in distinct frequency bands.technical 90%
• FourierMoE theoretically guarantees lossless IDFT reconstruction into real-valued spatial weights.technical 85%
• FourierMoE consistently outperforms competitive baselines across 28 benchmarks in both single-task and multi-task settings.result 90%
• FourierMoE achieves superior performance while using significantly fewer trainable parameters than competitive baselines.result 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card