{ "id": "rp-agentic-perception", "schema_version": "0.2", "type": "ai_native_research_lineage", "title": "具備自主感知策略的多模態智能架構系列", "title_en": "Agentic Perception: Toward Multimodal Intelligence with Autonomous Perception Strategy", "short_title": "APR", "status": "completed", "open_ended": false, "started_at": "2026-08-11", "completed_at": "2026-08-11", "reference_method": "8 篇論文的完整序列，篇序取自系列自身的檔名編號或內文自陳順序，非事後推斷。", "canonical_route": "/ai/programs/agentic-perception/", "visibility": "public", "languages": [ "zh-Hant" ], "agency": { "initiative": "human", "primary_authorship": "collaborative", "research_direction": "negotiated", "continuation_control": "human" }, "contributors": [ { "name": "Neo.K", "type": "human", "roles": [ "problem_proposer", "continuation_decider", "curator", "co_author" ] } ], "foundation_artifacts": [], "applied_artifacts": [], "iterations": [ { "sequence": 1, "id": "lm-002437", "title": "從均勻感知到自適應感知閱讀：多模態智能的感知資源配置理論", "topic": "從均勻感知到自適應感知閱讀：多模態智能的感知資源配置理論", "iteration_type": "foundation", "is_checkpoint": false, "canonical_url": "/p/lm-002437/", "raw_url": "/raw/lm-002437.md", "api_url": "/api/papers/lm-002437.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 2, "id": "lm-002436", "title": "從像素差分到語義差分：連續多模態感知中的變化階層", "topic": "從像素差分到語義差分：連續多模態感知中的變化階層", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002436/", "raw_url": "/raw/lm-002436.md", "api_url": "/api/papers/lm-002436.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 3, "id": "lm-002440", "title": "視覺不是一種閱讀：多尺度視覺閱讀模式與重觀察理論", "topic": "視覺不是一種閱讀：多尺度視覺閱讀模式與重觀察理論", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002440/", "raw_url": "/raw/lm-002440.md", "api_url": "/api/papers/lm-002440.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 4, "id": "lm-002439", "title": "持續世界狀態與差分重觀察：從連續影片理解到持續感知智能", "topic": "持續世界狀態與差分重觀察：從連續影片理解到持續感知智能", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002439/", "raw_url": "/raw/lm-002439.md", "api_url": "/api/papers/lm-002439.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 5, "id": "lm-002438", "title": "感知預算分配：從固定視覺 Token 到動態資訊獲益最大化", "topic": "感知預算分配：從固定視覺 Token 到動態資訊獲益最大化", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002438/", "raw_url": "/raw/lm-002438.md", "api_url": "/api/papers/lm-002438.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 6, "id": "lm-002441", "title": "跨模態感知閱讀：文字、影像、影片與聲音的統一注意策略", "topic": "跨模態感知閱讀：文字、影像、影片與聲音的統一注意策略", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002441/", "raw_url": "/raw/lm-002441.md", "api_url": "/api/papers/lm-002441.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 7, "id": "lm-002435", "title": "具備自主感知策略的多模態智能架構_Agentic Perception", "topic": "具備自主感知策略的多模態智能架構_Agentic Perception", "iteration_type": "applied", "is_checkpoint": false, "canonical_url": "/p/lm-002435/", "raw_url": "/raw/lm-002435.md", "api_url": "/api/papers/lm-002435.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" }, { "sequence": 8, "id": "lm-002442", "title": "適度認知論_智能為什麼不應對每一個問題都使用最大思考深度_v0.1", "topic": "適度認知論_智能為什麼不應對每一個問題都使用最大思考深度_v0.1", "iteration_type": "synthesis", "is_checkpoint": true, "canonical_url": "/p/lm-002442/", "raw_url": "/raw/lm-002442.md", "api_url": "/api/papers/lm-002442.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" } ], "current_state": { "sequence": 8, "artifact_id": "lm-002442", "status": "completed", "phase_status": "closed_by_design", "declaration": "本系列從自適應感知資源配置出發，依序處理像素差分到語義差分的變化階層、多尺度視覺閱讀、連續影片理解中的持續世界狀態與差分重觀察、感知預算的動態分配、跨模態統一注意策略，最終在第七篇收斂為完整的 Agentic Perception 架構——具備自主感知策略的多模態智能。第八篇《適度認知論》將同一套「不應對每個問題都用最大資源」的原則從感知領域推廣到一般推理深度配置，是系列核心洞見的跨領域延伸。", "stop_reason": "第七篇明確以「具備自主感知策略的多模態智能架構」作為系列封頂，統合前六篇成果；第八篇是同一原則向推理深度領域的延伸應用，非感知架構本身的延續，故列為系列附篇而非否認其封頂地位。", "unresolved_gaps": [], "next_actions": [], "handoff_required_tasks": [], "stewardship": "第二篇一處跨行 \\text{} 內含原始換行、位於 \\boxed{} 內；第一篇一處控制位元組（LF）吃掉 \\neq 的反斜線；第八篇整份文件 100% 使用裸 \\[...\\]／\\(...\\) 而非 $/$$（58 個顯示區塊 + 13 個行內），且該篇另有 2 處 \\boxed{} 未閉合（分別位於全文開篇論點與結尾論點兩處）——以上均已於發布前修復。原始檔案作者／機構／日期欄位已由另一個對話流程預先補上。第四篇（持續世界狀態與差分重觀察）與既有 lm-002089（從狀態機到世界狀態機）、lm-002336（持續世界狀態：母AI如何一直醒著）在「持久狀態＋差分更新優於每輪重建」的論證上高度收斂但應用領域不同（多模態感知 vs. 通用Agent架構 vs. 系統常駐性），屬獨立收斂而非重複，建議交叉引用。第七篇的雙時間尺度（FastLoop/SlowLoop）架構亦與既有 lm-002061（先反應再思考）呼應。", "cross_references": [ { "id": "lm-002089", "relation": "independently_convergent", "note": "持久狀態+差分更新的論證高度收斂，應用領域不同" }, { "id": "lm-002336", "relation": "independently_convergent", "note": "同上" }, { "id": "lm-002061", "relation": "independently_convergent", "note": "雙時間尺度架構呼應，應用領域不同（感知 vs. 具身社交時序）" } ], "artifact": { "sequence": 8, "id": "lm-002442", "title": "適度認知論_智能為什麼不應對每一個問題都使用最大思考深度_v0.1", "topic": "適度認知論_智能為什麼不應對每一個問題都使用最大思考深度_v0.1", "iteration_type": "synthesis", "is_checkpoint": true, "canonical_url": "/p/lm-002442/", "raw_url": "/raw/lm-002442.md", "api_url": "/api/papers/lm-002442.json", "authorship": "collaborative", "month": "2026-08", "language": "zh-Hant" } }, "integrity": { "files_seen": 8, "files_imported": 8, "missing_sequences": [], "missing_note": "", "duplicate_resolved": [] }, "generated_at": "2026-09-10T13:30:06Z" }