Google Research Blog
Google Research Blog 是 Google Research 的官方技術部落格,發布 Google 在機器學習、自然語言處理、AI 系統架構等領域的最新研究成果,也涵蓋 agent memory、RAG 與 model governance 類主題。
2026-11-05 新文:Detecting signs of hypertension using wearable sensors(AI-PPG-ACC-HTN)
Google Research(與 Beth Israel Deaconess Medical Center、Harvard Medical School 合作;Nature 論文第一作者 Paolo Di Achille、末位 Ming-Zher Poh)發布 AI-PPG-ACC-HTN:以消費級手錶既有的 PPG+ACC 被動訊號、無需 cuff 校準偵測高血壓跡象。方法為兩階段 ML——self-supervised contrastive learning 訓練 1-D CNN waveform encoder(492,347 小時、2,109 名參與者未標註資料),再將 person-level mean-aggregated embedding 結合身高送入 regularized logistic regression。驗證為前瞻性多中心研究(美國五站點、196 名無已知心血管疾病且未服血壓藥參與者,以 24h ABPM 為 gold standard):initial screening sensitivity/specificity/PPV=65.8%/90.0%/80.6%,對照 office screening 55.3%/—/77.8%(Nature abstract 同向支持並給 95% CI);confirmatory scenario(N=48)78.4%/90.9%/96.7% 維持部落格第一方口徑。來源自我限定為早期可行性研究、不取代標準臂帶或單獨確診,並稱結果影響了 Google Health app 的 blood pressure trends 功能;Nature 論文全文本輪未讀取(只讀頁面 metadata 與 abstract),相關 raw evidence:[[raw/articles/google-research-hypertension-wearable-sensors-2026-11-05]];可與 wearable-health-foundation-models、eval-is-spec 交叉閱讀。本輪只新增 raw evidence,未更新 compiled concept。
2026-11-04 新文:Agile Deliberation — AI 幫人類定義灰色地帶
Google Research(Otilia Stretcu、Leijie Wang)介紹 Agile Deliberation human-in-the-loop 框架:把主觀視覺 classifier(healthy food、inspiring artwork、內容審查 clickbait 圖等)的建立視為迭代對話,分 concept scoping(拆解 sub-concept hierarchy+explicit exclusions)與 concept iteration(VLM 主動挖掘 borderline images、使用者回饋餵入自動 prompt optimization loop)兩階段。評估為 18 場 × 1.5 小時 live user sessions(Healthy Food、Pay-to-Play),對照 zero-shot/modeling collaborator/manual deliberation 三 baseline;來源報告標註一致性顯著提高、classifier substantially outperform all standard baselines、effort/stress/mental demand 統計顯著下降且每位參與者偏好 Agile Deliberation。arXiv:2512.10821(CVPR 2026)abstract 第一方口徑支持 F1 較 automated decomposition baseline 高 7.5%、較 manual deliberation 高超過 3%;論文 PDF 全文、GitHub repo 與 Colab demo 本輪未讀取或執行,不視為取代人工標註或可直接生產部署的證據。相關 raw evidence:[[raw/articles/google-research-agile-deliberation-gray-areas-2026-11-04]];可與 eval-is-spec、user-simulator-evaluation 與 agentic-engineering-patterns 交叉閱讀。本輪只新增 raw evidence,未更新 compiled concept。
2026-10-27 新文:The sleep framework 與 continual learning
Google Research 提出 sleep framework,讓 LLM 在 active/wake 階段接收資訊,再於 offline sleep 階段進行 memory consolidation 與 dreaming,以把短期 context 轉成較穩定的參數記憶,並處理 catastrophic forgetting 與固定 knowledge cutoff。文章中的 benchmark 數字、10M tokens 與跨任務效果保留官方研究文章歸屬;arXiv 論文、完整實驗表與程式碼本輪未直接核對。相關 raw evidence:[[raw/articles/google-research-sleep-framework-continual-learning-2026-10-27]]。本輪只新增 raw evidence,未更新 compiled concept。
2026-10-23 新文:TB 篩檢前線的 foundation models
Google Research 回顧 AI 輔助胸腔 X 光 TB 篩檢從 benchmark、跨族群評估與前瞻性驗證,延伸到礦區等偏遠場域的 offline AI 部署,並介紹 HAI-DEF、MedSigLIP 與 MedGemma 作為可調整的開放 health foundation models。來源把系統定位為篩檢與後續確認檢測的入口,不支持降低死亡率、取代放射科醫師、單獨確診或直接全球部署;本輪只新增 raw evidence,未更新 compiled concept。相關 raw evidence:[[raw/articles/google-research-tb-screening-foundational-models-2026-10-23]]。
基本資訊
- 類型:官方研究部落格
- 營運方:Google Research / Google DeepMind
- 網站:https://research.google/blog/
- 語言:英文
2026-10-21 新文:AI-assisted ultrasound 與前線孕產照護
Google Research(Angelica Willis、Nichole Young-Lin)報告一項前瞻性多中心研究:讓非超音波專業操作員以低成本手持探頭執行 standardized blind sweep,再由 AI 估算胎齡。研究在芝加哥 Northwestern Medicine 與肯亞奈洛比 Jacaranda Health 測試;主要 cohort 385 名、16–36 週,來源報告 AI MAE 4.18 天,專家標準 4.48 天,芝加哥 adapted model 4.10 天,奈洛比在無當地 fine-tuning 下 4.27 天,並宣稱達臨床非劣性。這些數字保留 Google Research 對 JAMA Network Open 研究的轉述邊界;JAMA 全文與 supplement 本輪未取得,未視為 AI Ark 已獨立重算或臨床核准證據。來源亦記錄可替換電池、低品質 sweep rejection、confidence estimate 與不確定性呈現等部署/安全考量;不支持降低孕產婦死亡率或直接全球部署。相關 raw evidence:[[raw/articles/google-research-frontline-maternal-care-ai-assisted-ultrasound-2026-10-21]];可與 wearable-health-foundation-models、eval-is-spec 與 world-models 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-16 新文:Grounding AI agents in biological reality(autonomous wet lab building blocks)
Google Research(Pooja Rao、Petar Sirkovic、Annalisa Pawlosky)以三個研究項目展示把 AI agent 接地到實體濕實驗室的 building blocks:VERID(VERified Infrastructure for Discovery Laboratories)是建在開源 PyLabRobot 上的 semantic digital twin,以 live 3D state tracking、fast virtual dry runs、hardware safety checks 與 synchronized video/action logs 讓 agent 在物理安全約束下規劃、模擬並執行液體處理 protocol;aptamer lab-in-the-loop sandbox 以 7–10 天的 DNA aptamer design-test-analyze 閉環,讓 agent 從濕實驗室 readout 學習(myoglobin 33.5%/α-thrombin 42.1% on-target hit rate,agent 在 myoglobin Round 3 自主修訂 G-quadruplex 設計規則);agent-guided epitope excision 以局部 binding pocket 切除把 de novo binder folding 加速超過八倍,human–agent collaboration 在等 compute head-to-head 中幾乎翻倍 in-silico hit rate,並在 FAP target 上發現單數位納莫爾親和力的 binders。這些是 Google Research 研究 protocol 的來源陳述,三篇連結論文 PDF(VERID semantic lab report、agentic aptamer sandbox paper、epitope excision paper)全文、完整實驗設定與 KD 測量方法本輪未讀取或驗證,不視為 fully autonomous wet lab 已可部署或 binder 具藥物價值的證據。相關 raw evidence:[[raw/articles/google-research-grounding-ai-agents-biological-reality-autonomous-wet-labs-2026-10-16]];可與 academic-workflow-agents、chain-of-evidence-autonomous-research、data-science-agents 與 eval-is-spec 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-14 新文:Research@ 2026 十項研究重點 roundup
Google Research GM Yossi Matias 於 2026-10-14 Mountain View Research@ 活動發布十項研究重點:Earth AI early access 擴大(Ask Google Earth、Planetary Prediction Engine、Geospatial Reasoning agent A2A API,並整合 Dataset Search 與 TimesFM)、科學 multi-agent orchestration(流體力學 inverse design 用於導管設計案例)、Deep Review 公開化(早期版本已審查超過 16,000 份頂尖 CS 會議投稿)、Synaptics 以 Google open source IP 共同設計的新一代低功耗 edge chip、TimesFM/TabFM 宣布整合進 Gemini for business、MedGemma 加入 tool use 並探索新模態(Nature Medicine 發表後 >1,000 萬次下載)、AMIE The Lancet BIDMC 98 患者研究與 Included Health 全國性 randomized study、Groundsource 以 Gemini 建立超過 260 萬筆 flood events dataset 訓練 flash flood 預測模型、learning interactives(>30 個經教師審核的 AI 生成 STEM 教材)與 Play with Putty 多人協作原型工具;另宣布 Google.org 2M)。這些是官方 roundup post 的來源陳述,arXiv:2610.15750、Nature Medicine MedGemma 論文、The Lancet AMIE 研究全文、Synaptics 晶片公告與各產品操作本輪未讀取或驗證,十項能力維持 Google Research attribution/observational。相關 raw evidence:[[raw/articles/google-research-research-at-2026-amplifying-human-ingenuity-2026-10-14]];可與 planetary-prediction-engine、time-series-foundation-models、tabular-foundation-models 與 groundsource-news-to-data 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-14 新文:Deep Review(長篇技術文件的 agentic 驗證 pipeline)
Google Research(Rajesh Jayaram、Vincent Cohen-Addad)宣布 Deep Review 公開上線:由實驗性的 Paper Assistant Tool(PAT)演化而來的 agentic system,以四階段 pipeline(document segmentation、adaptive budgeting、deep parallel verification、global synthesis and agentic grounding)對長篇技術文件做邏輯一致性、數學陳述檢查與事實 grounding,並從學術論文審查擴展到設計文件、簡報與法律/財務文件。來源報告 PAT 已在 STOC 2026、ICML 2026、NeurIPS 2026、ICLR 2027 以 opt-in 方式為超過 16,000 份稿件提供預審回饋,並宣布每位使用者每天兩份免費審查、Gemini API key 分級用量與 Deep Review API;上傳文件不用於訓練且僅儲存一年。這些是 Google Research 的來源陳述與產品公告,arXiv:2606.28277〈Towards Automating Scientific Review with Google’s Paper Assistant Tool〉全文、pilot 統計表逐項數字、程式碼與產品操作本輪未讀取或驗證,不視為審查品質已量化驗證的證據。相關 raw evidence:[[raw/articles/google-research-deep-review-document-analysis-2026-10-14]];可與 academic-workflow-agents、chain-of-evidence-autonomous-research 與 eval-is-spec 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-14 新文:Geospatial Reasoning agent、Earth AI 與 planetary prediction
Google Research(Shravya Shetty、Luke Barrington)介紹實驗性的 Geospatial Reasoning agent:把 Earth AI foundation models(WeatherNext、Flood Forecasting、PDFM、AlphaEarth Foundations、Remote Sensing Foundations)、spatial tools(Maps Platform、Earth Engine、Data Commons、Dataset Search)、upfront planning 與 context 外 dedicated workspace 組合,支援多步驟空間分析、Planetary Prediction 與 BYOD API;技術報告比較四種 agent configuration 與四個 benchmark(CBGB、UnivEARTH、GeoExpertQueries、GeoKTools),並描述 WHO AFRO/DRC INRB、People in Need、GiveDirectly、WPP 與 Milken Institute 的早期測試。Ask Google Earth 與 headless A2A Agent API 開放 limited early access;12% prediction improvement、8X finer resolution、48 settlements、45,500 at-risk people 等數字維持 Google Research attribution/observational,arXiv:2610.15750 全文、benchmark payload、合作方原始資料與產品操作未讀取。相關 raw evidence:[[raw/articles/google-research-geospatial-reasoning-agent-earth-ai-2026-10-14]];可與 planetary-prediction-engine、data-science-agents 與 world-models 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
相關頁面
- empirical-research-assistance — ERA 與其 follow-up 應用案例
- academic-workflow-agents — PaperVizAgent / ScholarPeer 的學術工作流自動化
- chain-of-evidence-autonomous-research — Chain-of-Evidence、Science One 與研究 artifact 的可驗證性
- test-time-compute-evaluation — TTD-DR 的 draft-first、retrieval denoising 與 test-time budget 評估
- agentic-rag-cross-corpus — Google 提出的 Agentic RAG 架構(Sufficient Context Agent + Cross-Corpus Retrieval)
- machine-unlearning-audit — Google Research 的 machine unlearning 稽核方法
- reasoningbank-agent-memory — ReasoningBank:把成功與失敗經驗壓成可重用的 agent memory
- reasoning-for-factual-recall — reasoning trace 透過 computational buffer / factual priming 擴大 factual recall 邊界
- knowledge-profiling-factuality — 用 WikiProfile 拆解 encoding、recall、recognition 與 thinking recovery
- frozen-multi-token-prediction — Gemini Nano frozen MTP 手機端推論加速
- groundsource-news-to-data — Gemini 把多語新聞轉成結構化事件資料的 Groundsource 方法
- tabular-foundation-models — TabFM 與表格資料的 zero-shot foundation model 工作流
- time-series-foundation-models — TimesFM-3 與多變量時間序列、covariate、quantile forecast
- wearable-health-foundation-models — SensorFM 與 wearable sensor foundation model、agentic classroom、Personal Health Agent grounding
- world-models — 從靜態標籤與文字表徵延伸到 physical world 的狀態、功能與動態節律
- eval-is-spec — SymptomAI 的真實使用者、主動追問與延遲 ground truth 評測案例
- data-science-agents — DS-STAR 與資料科學 agent 的異質資料 / verifier loop 模式
- planetary-prediction-engine — PPE:LLM 編排 geospatial data selection、curation、AutoML 與 prediction
- test-time-memorization — Titans + MIRAS 的模型層長期記憶與 surprise-based update
- nested-learning-continual-learning — Nested Learning 的 continual learning / nested optimization 設計
- bayesian-teaching-llm-reasoning — 用 Bayesian assistant traces 訓練 LLM 的偏好推理與信念更新
- on-device-intent-extraction — 用小型 multimodal model 從 UI 操作軌跡抽取使用者意圖
- generative-ui — 模型依 prompt 生成完整互動介面的產品形態
- diffusion-model-creativity — score smoothing、data manifold 與 diffusion model 的生成新穎性
- sequential-attention-subset-selection — 用 attention score 做大型模型 subset selection 與 structured pruning
- privacy-preserving-chatbot-analytics — 用 differential privacy 做 chatbot 使用趨勢與 trace 聚合分析
- gist-smart-sampling — 用 Greedy Independent Set Thresholding 做訓練資料 smart sampling
- multilingual-model-scaling-laws — ATLAS 多語言模型 scaling laws 與 cross-lingual transfer 判準
- google-ai — Google 官方 X 帳號與相關發布來源
2026-10-09 新文:Does better work always mean better workers?(AI 與專業技能形成的三個月 RCT)
Google Research(David Autor、Tanya Rodchenko,NBER w35720 共同作者)以 133 位專利律師、11 家 IP 律所的三個月隨機對照試驗報告:AI 專利寫作工具在撰寫任務上把分數提高 0.34→0.38 SD(約 10–11 個百分位)、初級律師伴隨顯著時間節省;但在禁止使用 AI 的 redlining 判斷力測試上,改善完全由資深律師驅動(+0.45 SD/約 13 個百分位),初級律師無平均技能增益且分數兩極化。作者把「提升表現」與「建立持久判斷力」視為可能互相張力的目標;這些是 Google Research 研究 protocol 的來源陳述,NBER w35720 論文全文、完整統計表與所連結文獻原文未讀取,不視為可外推到其他職業或更長時期的證據。相關 raw evidence:[[raw/articles/google-research-does-better-work-always-mean-better-workers-2026-10-09]];可與 ai-collaboration-compounding-workflow、agentic-ai-cost-management 及 AI 工作/職涯類記錄交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-06 新文:Earth AI planetary geospatial foundation models 與全球公共衛生
Google Research 以 Population Dynamics Foundation Model(PDFM)示範 planetary geospatial foundation model 如何接入既有公共衛生與流行病學 workflow,整理五個 partner-driven 案例:美加邊境 MMR、美國 CVD nowcasting、墨西哥 dengue、CDC PRAMS 產後憂鬱與剛果民主共和國 cholera。來源報告跨境資訊 spillover、資料新鮮度、短期預測與 4–8 週 outbreak warning 的量化改善或可比表現;這些數字仍是 Google Research 研究 protocol 的來源陳述,完整論文、原始資料、程式碼與獨立重現未讀取,不視為已通過公共衛生 operational deployment 的證據。相關 raw evidence:[[raw/articles/google-research-earth-ai-public-health-2026-10-06]];可與 planetary-prediction-engine、world-models、wearable-health-foundation-models、data-science-agents 與 empirical-research-assistance 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-05 新文:Agentic Privacy and Security 的 Contextual Integrity 視角
Google Research 以 Contextual Integrity 為鏡頭,整理 agentic privacy/security 的三個結構性挑戰:輸入歧義與 prompt injection、機率式控制流程,以及自主性/委派造成的監督失效與 confirmation fatigue;並提出 contextual policy engine、多層防禦、multi-agent guardrails 與動態 Agent Gym 評測作為研究議程。這是官方 workshop report 的研究問題整理,不代表已有通用、可靠或 production-ready 的控制器。相關 raw evidence:[[raw/articles/google-research-agentic-privacy-security-contextual-integrity-2026-10-05]];可與 identity-aware-connectivity-platform、ai-native-enterprise-governance、agent-trace-observability 與 privacy-preserving-chatbot-analytics 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-10-02 新文:Toward provably private learning from federated data
Google Research 介紹以 TEEs 為基礎的下一代 Federated Learning 系統:client device 本地加密上傳+預授權 access policy、TEE-hosted KMS(RAFT consensus)只對符合 policy 的 workload 釋放金鑰、Federated Language 表達的 server-side training loop(root/worker TEEs)、KMS-encrypted fault-tolerant recovery,以及 Rekor public transparency log 與 reproducible builds。Gboard English/Japanese next-word prediction 已 production deployment;DP epsilon/delta 數值、privacy-utility curve 完整實驗設定、Rekor log entries 與 reproducible build 流程本輪未獨立稽核或讀取,維持 Google Research/arXiv:2609.31494 attribution。相關 raw evidence:[[raw/articles/google-research-provably-private-federated-learning-tee-2026-10-02]];可與 privacy-preserving-chatbot-analytics 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-09-29 新文:Diffusion Controller
Google Research 介紹 Diffusion Controller(DiffCon):把 reverse diffusion denoising 視為連續控制問題,在 frozen image-generation backbone 上接輕量 side network,透過 policy-gradient/PPO 或 reward-weighted regression 做 prompt/preference alignment。文章報告 Stable Diffusion v1.4、HPS-v2 與 white-box/gray-box 比較結果;90% win rate、LoRA 對照與「不破壞穩定性」均保留為 Google Research/arXiv 研究 attribution,本輪未讀取完整論文、實驗表格、模型權重或重跑 benchmark。相關 raw evidence:[[raw/articles/google-research-diffusion-controller-2026-09-29]];可與 diffusion-model-creativity、generative-media-agent-workflow 交叉閱讀。本輪只新增 raw evidence,不更新 compiled concept。
2026-09-24 新文:Automating coherent long-form video generation
Google Research 介紹 AI video co-director、CANVAS、A²RD 與 VQQA 四個互相連接的研究框架,把長篇影片生成拆成全域創意搜尋、persistent visual/video memory、segment-level retrieve-synthesize-refine-update 與 VLM closed-loop selection。文章把目標放在降低 semantic drift、cascading failures、feature drift 與 content collapse,並以 GenAD-Bench、HardContinuityBench、VBench-Long 等 benchmark 報告研究框架下的改善;AI Ark 只保留 Google Research 的來源陳述,未讀取連結的 ArXiv paper、benchmark、code 或模型,未做獨立重跑。相關 raw evidence:[[raw/articles/google-research-coherent-long-form-video-generation-2026-09-24]];可與 generative-media-agent-workflow、dynamic-agent-workflows、agent-trace-observability 與 eval-is-spec 交叉閱讀。
2026-09-15 新文:Retrieve-for-Train(R4T)
Google Research 發布 R4T(Retrieve-for-Train),一個以 RL 作為一次性 objective transducer 的 set-valued retrieval framework:先以 Soft-GRPO + composite set-level reward(groundedness、Vendi Score diversity、alignment 三項互為 counter-anchor)訓練 Gemma3-4B/Qwen3-4B fan-out LLM,離線合成 (query → target-set) supervision pairs,再 distill 成 53.9M-parameter diffusion retriever,在 continuous embedding space 以單一 non-autoregressive pass 生成完整 sub-query set。來源在 fashion(CLIP)與 music playlist(MuLan)benchmark 報告優於 single-query、zero-shot expansion 與 Best-of-N baseline,並宣稱 12–20× latency speedup over autoregressive fan-out;這些是 ICML 2026 論文(arXiv:2603.06397)研究 protocol 下的來源陳述,不是 AI Ark 的獨立重現或跨 domain/production 證據。相關 raw evidence:[[raw/articles/google-research-retrieve-for-train-rl-compiled-diffusion-fan-out-retrieval-2026-09-15]];可與 agentic-rag-cross-corpus、document-parsing-first-rag、test-time-compute-evaluation 交叉閱讀。
2026-09-10 新文:ToolGrad
Google Research 發布 ToolGrad,提出 answer-first tool-use dataset generation:先以 textual gradients 引導 API workflow 的提案、執行、選擇與更新,再生成相符的 user query。來源以 ToolBench 與 BFCL 報告較高的資料生成 pass rate、較低成本與 ToolGrad-12B 的 tool-use benchmark 結果;這些是研究 protocol 與 benchmark snapshot 下的來源陳述,不是 AI Ark 的獨立重現或跨版本模型排名。相關 raw evidence:[[raw/articles/google-research-toolgrad-tool-use-dataset-generation-2026-09-10]];可與 agent-post-training-workflow、dynamic-agent-workflows、eval-is-spec 交叉閱讀。
2026-09-03 新文:跨族群 genomic prediction
Google Research 介紹以 UK Biobank 歐洲樣本與 Biobank Japan 日本樣本評估 polygenic risk score(PRS)transfer learning 的研究,跨八個 trait 改變 source/target sample size,比較 UKB discovery + elastic net、cross-population meta-analysis 與 PRS-CSx。來源報告小型 target cohort 可受外部資料 pooling 幫助,但 target cohort 增大後,尤其是較 population-specific 的 trait,target-only training 可能表現更好;這些 sample-size crossover 是特定研究 protocol 的結果,不是臨床或其他族群的固定門檻。相關 raw evidence:[[raw/articles/google-research-transfer-learning-genomic-prediction-2026-09-03]];可與 wearable-health-foundation-models、data-science-agents 和 eval-is-spec 交叉閱讀。
2026-08-26 新文:GlucoFM
Google Research 發布 GlucoFM,一個針對 continuous glucose monitoring(CGM)的 lightweight self-supervised foundation model。它用 dual-stream encoder 分開較慢的 glucose trend 與短期 deviation,保留 time-of-day、missingness 與 sensor variation,並在四個 cohort、七項臨床預測任務、PPGR forecasting、跨 cohort transfer 與 few-shot 設定中評估 frozen representation;這些結果屬研究資料與來源 protocol 下的 attribution,不等於臨床診斷或部署安全 benchmark。相關知識連結至 wearable-health-foundation-models、data-science-agents 與 eval-is-spec。
2026-08-27 新文:Planetary Prediction Engine(PPE)
Google Research 發布 Planetary Prediction Engine(PPE),把自然語言 geospatial query 交給 LLM 編排的 intelligent data selection、dataset curation、AutoML & prediction 流程,並以 opaque handles 在階段間傳遞大型 data artifacts。來源在美國公共衛生、奈及利亞 food security 與 DRC Ebola nowcasting 報告多項 benchmark 改善;這些結果仍屬 Google Research 的 early-stage experimental capability 與來源 attribution,不視為獨立重現或可直接部署的證據。相關知識連結至 planetary-prediction-engine、data-science-agents、world-models 與 empirical-research-assistance。
2026-08-31 新文:TimesFM-3
Google Research 發布 TimesFM-3,主打原生 multivariate time-series forecasting、multiple targets、past/past-future covariates、single-forward-pass 解碼與 point/quantile outputs。來源稱模型以 330M parameters、超過 1 trillion time points 的 real-world/synthetic corpus 預訓練,並在 Gift-Eval、FEV-Bench 與 TIME 報告最高的 foundation-model average rank;官方 GitHub README 另列 FEV-Bench、TIME 與 GIFT-Eval 的 rank #1 說法。這些是 Google Research 與官方 repo 的研究/benchmark attribution,不是 AI Ark 的獨立重現或 production SOTA。相關知識連結至 time-series-foundation-models、tabular-foundation-models、data-science-agents 與 eval-is-spec。
2026-09-01 新文:MAPL-EMIT
Google Research 發布 MAPL-EMIT,使用 Swin-S vision transformer、EMIT hyperspectral radiance 與 spatial context,自動進行 methane plume detection、enhancement quantification、delineation 與 source localization。文章與 arXiv 摘要報告在 1,084 個 EMIT granules 上捕捉 84% NASA L2B expert-annotated plume complexes,並找到約 1.5 倍 human analysts 的 plausible plumes;這些數字屬研究 protocol 下的來源陳述,不是獨立重現或減排成效 benchmark。相關知識連結至 world-models。
2026-08-25 新文:AgentHands
Google Research 發布 AgentHands XR prototype:以 scene registry、GestureEvents、文字/語音時間戳與本地 headset parser,把 LLM 回應轉成空間定位、動作示範與安全提示。來源報告 N = 12 的 within-subjects study,相較 speech-only baseline 在物件定位、複雜動作理解、警告注意力與認知負荷上有改善;這些結果屬 Google Research 的小型研究原型與特定任務研究,不視為通用 usability 或部署安全 benchmark。相關知識連結至 agent-experience、generative-ui、natively-adaptive-interfaces 與 agent-event-streaming-format。
Raw Sources
raw/articles/google-research-hypertension-wearable-sensors-2026-11-05.md— Detecting signs of hypertension using wearable sensors(AI-PPG-ACC-HTN:手錶 PPG+ACC 被動訊號、無 cuff 校準偵測高血壓跡象;Nature DOI 10.1038/s41586-026-11336-3)raw/articles/google-research-does-better-work-always-mean-better-workers-2026-10-09.md— Does better work always mean better workers?(三個月專利律師 RCT:AI 提升撰寫品質,但移除 AI 後的判斷力改善完全由資深律師驅動)raw/articles/google-research-provably-private-federated-learning-tee-2026-10-02.md— Toward provably private learning from federated data(TEE-based FL、KMS/Rekor/Federated Language 與 Gboard deployment)raw/articles/google-research-millemiglia-middle-mile-logistics-instance-generator-2026-09-18.md— MilleMiglia:middle-mile logistics instance generator(C++/Protocol Buffers、space-time graph multi-commodity flow、fixed schedules/throughput/synchronization constraints)raw/articles/google-research-agentic-rag-2026-06-05.md— Unlocking Dependable Responses with Gemini Enterprise Agent Platform’s Agentic RAGraw/articles/google-research-machine-unlearning-audit-2026-06-10.md— New framework for auditing machine unlearningraw/articles/google-research-reasoningbank-agent-memory-2026-04-21.md— ReasoningBank: Enabling agents to learn from experienceraw/articles/google-research-era-scientist-usage-2026-04-29.md— Four ways Google Research scientists have been using Empirical Research Assistanceraw/articles/google-research-alphaevolve-theoretical-cs-2025-09-30.md— AI as a research partner: Advancing theoretical computer science with AlphaEvolveraw/articles/google-research-science-one-chain-of-evidence-2026-07-30.md— Science One Framework: A verifiable autonomous research framework via Chain-of-Evidenceraw/articles/google-research-deep-researcher-test-time-diffusion-2025-09-19.md— Deep researcher with test-time diffusionraw/articles/google-research-thinking-to-recall-parametric-knowledge-2026-06-24.md— Thinking to recall: How reasoning unlocks parametric knowledge in LLMsraw/articles/google-research-gemini-nano-frozen-mtp-2026-06-26.md— Accelerating Gemini Nano models on Pixel with frozen Multi-Token Predictionraw/articles/google-research-groundsource-news-to-data-gemini-2026-03-12.md— Introducing Groundsource: Turning news reports into data with Geminiraw/articles/google-research-tabfm-zero-shot-tabular-data-2026-06-30.md— Introducing TabFM: A zero-shot foundation model for tabular dataraw/articles/google-research-sensorfm-wearable-health-data-2026-07-09.md— SensorFM: Towards a general intelligence and interface for wearable health dataraw/articles/google-research-symptomai-everyday-symptom-assessment-2026-07-22.md— SymptomAI: Towards a conversational AI agent for everyday symptom assessmentraw/articles/google-research-amie-audio-visual-clinical-consultations-2026-08-11.md— AMIE (Video):real-time audio-visual clinical consultation、三 agent 架構與 randomized OSCEraw/articles/google-research-knowledge-profiling-parametric-factuality-2026-08-12.md— Empty shelves or lost keys? Recall is the bottleneck for parametric factualityraw/articles/google-research-photoscan-cardiometabolic-risk-2026-08-17.md— PhotoScan:用 smartphone imagery 估計 body composition 與 cardiometabolic riskraw/articles/google-research-biomarker-discovery-framework-2026-08-21.md— Biomarker Discovery Framework:wearable biomarker prioritization、adversarial multi-agent loop 與 human expert evaluationraw/articles/google-research-mobility-embedded-pois-2026-08-21.md— Mobility-Embedded POIs、文字與匿名 mobility 融合、場所功能節律與 unseen-place evaluationraw/articles/google-research-agenthands-spatially-grounded-agent-conversations-xr-2026-08-25.md— AgentHands:XR 空間定位、GestureEvents、同步手勢與 N=12 使用者研究raw/articles/google-research-glucofm-continuous-glucose-monitoring-2026-08-26.md— GlucoFM:CGM dual-stream foundation model、latent prediction、跨 cohort transfer 與 few-shot evaluationraw/articles/google-research-planetary-prediction-engine-earth-ai-2026-08-27.md— Planetary Prediction Engine:geospatial data selection、dataset curation、AutoML、multimodal fusion 與 prediction benchmarksraw/articles/google-research-timesfm-3-multivariate-forecasting-2026-08-31.md— TimesFM-3:多變量時間序列、covariates、single-pass quantile forecasting 與官方 benchmark/授權邊界raw/articles/google-research-mapl-emit-methane-deep-learning-2026-09-01.md— MAPL-EMIT:hyperspectral methane plume 偵測、source localization、synthetic training、benchmark 與信心/物理單位邊界raw/articles/google-research-ds-star-data-science-agent-2025-11-06.md— DS-STAR: A state-of-the-art versatile data science agentraw/articles/google-research-dialoglab-human-ai-group-conversations-2026-02-10.md— Beyond one-on-one: Authoring, simulating, and testing dynamic human-AI group conversationsraw/articles/google-research-titans-miras-long-term-memory-2025-12-04.md— Titans + MIRAS: Helping AI have long-term memoryraw/articles/google-research-bayesian-teaching-llm-reasoning-2026-03-04.md— Teaching LLMs to reason like Bayesiansraw/articles/google-research-nested-learning-continual-learning-2025-11-07.md— Introducing Nested Learning: A new ML paradigm for continual learningraw/articles/google-research-small-models-intent-extraction-decomposition-2026-01-22.md— Small models, big results: Achieving superior intent extraction through decompositionraw/articles/google-research-generative-ui-2025-11-18.md— Generative UI: A rich, custom, visual interactive user experience for any promptraw/articles/google-research-sequential-attention-model-efficiency-2026-02-04.md— Sequential Attention: Making AI models leaner and faster without sacrificing accuracyraw/articles/google-research-differentially-private-chatbot-use-2025-12-10.md— A differentially private framework for gaining insights into AI chatbot useraw/articles/google-research-gist-smart-sampling-2026-01-23.md— Introducing GIST: The next stage in smart samplingraw/articles/google-research-atlas-multilingual-scaling-laws-2026-01-27.md— ATLAS: Practical scaling laws for multilingual modelsraw/articles/google-research-toolgrad-tool-use-dataset-generation-2026-09-10.md— ToolGrad: Efficient tool-use dataset generation with textual “gradients”raw/articles/google-research-retrieve-for-train-rl-compiled-diffusion-fan-out-retrieval-2026-09-15.md— Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train(R4T)