Find a story
Search Spins
Search titles, summaries, and missing voices across published articles — press releases, announcements, and media coverage.
0 results for “model compression”
Progressive$^2$: A Teacher-Student Progressive Co-Evolving Knowledge Distillation Method for Substantial Model Compression
A new knowledge distillation method called Progressive$^2$ is introduced to improve model compression by enabling co-evolution of teacher and student models through progressive layer selection and iterative size reduction.
Aug 4, 2026
Keyword Matters: Unveiling the Energy Sensitivity of On-Device LLM Prompting
A new arXiv preprint presents empirical evidence that prompt wording—especially imperative verbs and instruction structure—affects energy consumption during on-device LLM inference, revealing a previously underexplored optimization lever.
Jul 28, 2026
Break Through the Compression Bottleneck: From Theory to Practice
A new arXiv paper identifies a previously unrecognized non-orthogonality between low-rank decomposition and quantization—two core LLM compression techniques—and introduces Diagonal Adhesive Method (DAM) to mitigate resulting performance degradation.
Jul 24, 2026
1stProtect and Multiverse Computing Partner to Deliver Secure AI Inference at the Edge -- No Cloud Required
1stProtect and Multiverse Computing announced a partnership to deliver on-device AI inference using quantum-inspired model compression and real-time runtime enforcement, claiming enhanced security and performance without cloud dependency.
Published Jul 3, 2026 · Analyzed Jul 6, 2026