Papers on skulldchttps://skulldc.net/categories/papers/Recent content in Papers on skulldcHugoen-usFri, 06 Nov 2026 10:26:06 +0000DPDM routes agent decisions through one frozen conditional predictor instead of promptinghttps://skulldc.net/2026/11/dpdm-routes-agent-decisions-through-one-frozen-conditional-predictor-i/Fri, 06 Nov 2026 10:26:06 +0000https://skulldc.net/2026/11/dpdm-routes-agent-decisions-through-one-frozen-conditional-predictor-i/A paper proposes DPDM, which pairs a language model encoder with a frozen pretrained predictor so agents learn from deployment history without any weight updates.NEO-VLA drops the vision encoder so action policies can read visual features at any depthhttps://skulldc.net/2026/11/neo-vla-drops-the-vision-encoder-so-action-policies-can-read-visual-fe/Thu, 05 Nov 2026 10:25:39 +0000https://skulldc.net/2026/11/neo-vla-drops-the-vision-encoder-so-action-policies-can-read-visual-fe/A paper from MMLab@NTU argues that vision-language-action models lose action-critical detail in their visual encoders and proposes NEO-VLA, an encoder-free design.FactEmbed factorizes MLLM embeddings for composed image retrievalhttps://skulldc.net/2026/11/factembed-factorizes-mllm-embeddings-for-composed-image-retrieval/Wed, 04 Nov 2026 20:26:21 +0000https://skulldc.net/2026/11/factembed-factorizes-mllm-embeddings-for-composed-image-retrieval/A KAIST AI paper makes an MLLM imagine the requested image edit through factorized latents before building retrieval embeddings, and reuses those latents for cheap reranking.Route Anything studies when routing between models beats one fixed modelhttps://skulldc.net/2026/11/route-anything-studies-when-routing-between-models-beats-one-fixed-mod/Wed, 04 Nov 2026 10:27:47 +0000https://skulldc.net/2026/11/route-anything-studies-when-routing-between-models-beats-one-fixed-mod/A new paper builds a 374,401-run trajectory matrix over 25 benchmarks and 452 candidate configurations to test how far learned routers can beat a single fixed candidate.ToolSpace builds agent training environments from real MCP tool traceshttps://skulldc.net/2026/11/toolspace-builds-agent-training-environments-from-real-mcp-tool-traces/Wed, 04 Nov 2026 10:27:47 +0000https://skulldc.net/2026/11/toolspace-builds-agent-training-environments-from-real-mcp-tool-traces/ToolSpace mines tool dependencies from real MCP execution traces, turns them into environments and checked tasks, and reports large gains for open models on long-horizon tool use.Reviso trains a video captioning agent to choose which visual evidence to inspecthttps://skulldc.net/2026/11/reviso-trains-a-video-captioning-agent-to-choose-which-visual-evidence/Tue, 03 Nov 2026 10:27:42 +0000https://skulldc.net/2026/11/reviso-trains-a-video-captioning-agent-to-choose-which-visual-evidence/An agentic video captioner plans what to look at from a low-frame-rate preview, revisits key segments with tools, and is trained with GRPO and cost-aware rewards.ServiceNow-AI Releases Apriel Voice 27B Preview Open Speech-Language Modelhttps://skulldc.net/2026/11/servicenow-ai-releases-apriel-voice-27b-preview-open-speech-language-m/Tue, 03 Nov 2026 00:14:36 +0000https://skulldc.net/2026/11/servicenow-ai-releases-apriel-voice-27b-preview-open-speech-language-m/ServiceNow-AI introduces Apriel Voice 27B Preview, an open speech-language model built on Gemma 4 26B-A4B that unifies speech understanding, reasoning, and tool use with reported benchmarks.