[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nasa-ibm-lunar-foundation-model-sombench":3,"topics-all":42,"news-related-c99751d5-418e-49d5-99d3-e43b84c80ec7":60},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":28,"news_slug":35,"published_at":36,"created_at":37,"modified_at":38,"is_published":39,"publish_type":40,"image_url":15,"view_count":41},"c99751d5-418e-49d5-99d3-e43b84c80ec7","IBM与NASA开源月球基础模型:Lunar Foundation Model","9月10日,IBM与NASA联合开源NASA-IBM Lunar Foundation Model,首批公众可用的月球基础模型之一。极区冰沉积预测RMSE降到0.0293,优于SwinV2-B基线最多22%,配套SomBench数据集。","2026 年 9 月 10 日,IBM 与 NASA 联合开源了 NASA-IBM Lunar Foundation Model,这是首批面向月球科学、面向公众可用的基础模型之一。模型权重以 Apache-2.0 协议托管在 Hugging Face,微调代码与训练管线放在 NASA-IMPACT 的 GitHub 仓库,下游适配走 TerraTorch 框架。整套发布第一次把覆盖 17 年观测的 LRO 数据,加上 GRAIL 重力、JAXA SELENE\u002FKaguya 多源数据,聚合成了一个机器学习就绪的多模态、多分辨率月面数据集 SomBench,和模型一起同步开放给研究者。\n\n## 架构与预训练:从零训的 ViT-B,几何信息当一等公民\n\n模型走的是 ViT-B 编解码器架构:12 层、768 维、12 个注意力头,从头训练而非继承自 ImageNet。训练集是 SomBench 里约 200 万对空间配准的月球切片瓦片,涵盖 11 种模态,两种空间尺度——LRO 窄角相机约 1 米\u002F像素的窄角图像,以及宽角相机约 100 米\u002F像素的宽角图像,对应 NAC_HighRes 与 WAC_LowRes 两个数据轨道。其中窄角切片来自 1095 张配准后的 NAC 帧共 100 万余瓦片、约 1.4 TB;宽角切片来自 54080 条 WAC 记录共 96 万余瓦片、约 38 TB。两者在同一套权重下做混合 batch 训练,跨越 100 倍分辨率尺度仍共用同一套编码器,这是发布里被反复强调的一点。\n\n预训练规模是 16 张 H100、跑 15 万步、global batch 1536、bf16 精度,大约 1100 GPU-小时。模态编码用 9 个 VQ-VAE tokenizer,FSQ 量化,DDPM 解码,目标函数是离散词表上的交叉熵。每个瓦片的光照几何(太阳角、相机帧锚、瓦片足迹)被显式 token 化作为编码器输入——这一点很关键,官方在技术报告里直接点明:月球表面外观主要由光照几何决定,而不是本征反射率变化,所以把几何当作一等公民塞进输入,比把所有瓦片当 i.i.d. 图像处理要靠谱。\n\n## 基准表现:最大 margin 出现在极区冰沉积\n\n基准表现是这份发布里最有信息量的部分。官方给的核心数字是:在月球极区冰沉积预测任务上,模型 RMSE 降到 0.0293,而 SwinV2-B(ImageNet 预训练)在同一基准上是 0.0377,随机初始化对照组是 0.0397——这一项也是相对基线提升最大的地方,技术报告里把它标为「最宽 margin」。在不规则月海斑块(IMP)分割任务上,IoU1 跑到 0.5709,基线 ConvNeXtV2-B 是 0.5687,差距小但在带噪声标签的设定下仍维持优势。在 WAC 影像 Robbins 陨石坑检测上,用 LoRA 微调 mAP 跑到 0.2581,SwinV2-B 是 0.2420。IBM 的口径是「极地冰沉积探测误差降低最多 22%,撞击坑检测在 100 米上下文尺度上以一半训练数据超过 SwinV2-B 约 19%」——这两个最大幅度数字是同一份技术报告里的最大值,不是平均值,作者自己也在脚注里提示 NAC 陨石坑和 IMP 两项的领先 margin 接近种子间波动,应当视为「可比」而非显著优于。\n\n## 边界与定位:加速假设生成,不是最后一公里决策工具\n\n需要把边界写清楚。模型卡自承三点:不是科学级生成模型;没有自带大地测量参考框架;未经任务方验证用于着陆点认证或 hazard clearance 这种操作决策。冰沉积输出回归的是一个由先验知识构建的模糊叠合图,而不是实测冰。SomBench 数据卡也注明了异质空间分辨率与数十米级绝对地理定位不确定性。换句话说:这是给研究者加速假设生成、缩小搜索空间的预训练底座,不是给任务方当最后一公里决策工具。\n\n## Prithvi 系列的延展:从地球到月球的「基座 + 数据 + 工具链」三件套\n\n这套模型也是 IBM-NASA 合作的 Prithvi 开源基础模型家族的新成员,此前已经覆盖地理空间、天气、日球层物理,现在加上了月球。NASA 首席科学数据官 Kevin Murphy 与 IBM Research Europe UK\u002FIreland 负责人 Juan Bernabe-Moreno 的官方表态都强调「让科学家从共享基座出发,而不是每个问题重造一套算法」——这也是这一系列发布的一贯叙事。训练数据走的是 NASA 80MSFC25M0084 奖项下完成的工作。整套东西已经在 Hugging Face(`nasa-ibm-ai4science\u002FNASA-IBM-Lunar-Foundation-Model`)和 NASA-IMPACT 的 GitHub 上线。\n\n## 所以呢:为什么这次发布值得专门拆一篇\n\nPrithvi 系列把「领域基础模型 + 公开配准数据集」这套范式从地球延到了月球,意味着以后做月面研究的实验室不需要再从零训 ViT,也不需要再为对齐 LROC、GRAIL、Kaguya 多个分辨率档位的数据写胶水代码,直接 fine-tune 或 LoRA 适配到自己的细分任务。这种「数据集 + 预训练底座 + 下游工具链一起开源」的形态,放在 NASA 重返月球的背景下,实际上把月面遥感这件事从「少数团队能做的闭门工作」推进到了「任何地球上的 CV 组都可以上手」。这也是为什么 IBM 把它框定为「one of the first publicly available foundation models for scientific exploration of the Moon」——重点不是「最大」「最强」,而是「公开、可改、可用」。这一发布也部分回应了科学界对 AI for Science 工具能否真正开放的反复质疑:权重 Apache-2.0、代码 GitHub 开源、数据 CC BY 4.0 三件套同时到位,在同类地球科学模型发布里算是走得最远的一次。\n\n参考资料:\n- IBM newsroom 官方发布(主要来源):https:\u002F\u002Fnewsroom.ibm.com\u002F2026-09-10-ibm-and-nasa-release-open-source-ai-model-to-support-lunar-exploration\n- Unite.AI 独立技术解读(包含完整架构细节、benchmark 表与训练细节):https:\u002F\u002Fwww.unite.ai\u002Fibm-and-nasa-open-source-lunar-foundation-model-with-sombench-dataset\u002F","https:\u002F\u002Fnewsroom.ibm.com\u002F2026-09-10-ibm-and-nasa-release-open-source-ai-model-to-support-lunar-exploration","653dda08-2edc-4d17-aeb2-56b0c88dd918",[11,16,19,22,25],{"id":12,"name":13,"slug":13,"description":14,"color":15},"9112951a-2abb-4214-b63a-385ec7afb2ba","ai-for-science","AI for Science 专题：追踪 AI 在生命科学、化学材料、物理世界模型等科学方向的关键突破",null,{"id":17,"name":18,"slug":18,"description":15,"color":15},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":20,"name":21,"slug":21,"description":15,"color":15},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":23,"name":24,"slug":24,"description":15,"color":15},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",{"id":26,"name":27,"slug":27,"description":15,"color":15},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[29],{"id":30,"lang":31,"title":32,"summary":33,"content":34},"24237271-782a-4d83-8f15-ef689da462f4","en","IBM and NASA Open-Source the Lunar Foundation Model","IBM and NASA released the open-source NASA-IBM Lunar Foundation Model on Sept 10. Polar ice RMSE 0.0293 beats SwinV2-B by 22%. SomBench ships alongside.","On September 10, 2026, IBM and NASA jointly open-sourced the NASA-IBM Lunar Foundation Model, one of the first foundation models built specifically for lunar science and made publicly available. Model weights are released under Apache-2.0 on Hugging Face; fine-tuning code and training pipeline live in the NASA-IMPACT GitHub repository, with downstream adaptation routed through the TerraTorch framework. The release is paired with SomBench, the first open-source multi-modal, multi-resolution lunar dataset aggregating LRO's 17-year observation record with GRAIL gravity and JAXA SELENE\u002FKaguya supplementary data — an ML-ready corpus the lunar science community can build on directly.\n\n## Architecture and Pretraining: A ViT-B Trained From Scratch With Geometry as a First-Class Input\n\nThe architecture is a ViT-B encoder-decoder trained from scratch — 12 layers, 768 dimensions, 12 attention heads — not inherited from ImageNet weights. The training corpus is roughly 2 million spatially co-registered lunar tile bundles from SomBench, spanning 11 modalities at two spatial scales: LRO Narrow Angle Camera imagery at approximately 1 meter per pixel, and Wide Angle Camera imagery at approximately 100 meters per pixel, organized as NAC_HighRes and WAC_LowRes data tracks. The narrow-angle track draws from 1,095 co-registered NAC frames totaling just over 1 million tiles and approximately 1.4 TB; the wide-angle track pulls from 54,080 WAC records totaling just under 964,000 tiles and approximately 38 TB. Both tracks train together in a single mixed-batch loop with one shared set of weights across a 100× resolution gap — a design choice the release emphasizes.\n\nPretraining ran on 16 H100 GPUs for 150,000 steps at global batch 1,536 in bf16, totaling approximately 1,100 GPU-hours. Modality encoding uses nine modality-specific VQ-VAE tokenizers with FSQ quantization and a DDPM decoder; the objective is cross-entropy over discrete token vocabularies. Per-tile acquisition geometry — illumination angles, solar-frame anchors, tile footprint — is tokenized as explicit encoder inputs. The technical report states this directly: lunar surface appearance is governed more by illumination geometry than by intrinsic surface variation, so treating geometry as a first-class input is more faithful than treating all tiles as i.i.d. images.\n\n## Benchmark Results: The Widest Margin Is on Polar Ice Prospectivity\n\nThe benchmark table is the most informative part of the release. The headline numbers: on polar lunar ice prospectivity, the model reaches RMSE 0.0293, against 0.0377 for SwinV2-B (ImageNet pretrained) and 0.0397 for a random-initialization control — ice prospectivity is flagged in the report as the widest-margin task. On Irregular Mare Patch (IMP) segmentation, IoU1 hits 0.5709 versus 0.5687 for ConvNeXtV2-B; the margin is small but the model maintains the lead under noisy labels. On WAC Robbins crater detection with LoRA fine-tuning, mAP reaches 0.2581 against 0.2420 for SwinV2-B. IBM's published framing is \"up to 22% lower RMSE on ice prospectivity\" and \"nearly 19% better at roughly 100-meter context scale using half the training data\" — these are the maximum-margin figures from the same technical paper, not averages, and the authors themselves footnote that NAC crater and IMP margins are within seed-spread and should be read as \"comparable\" rather than significantly better.\n\n## Boundaries and Positioning: Accelerating Hypothesis Generation, Not Last-Mile Mission Decisions\n\nThe boundaries need to be explicit. The model card self-states three limits: it is not a science-grade generative model; it carries no geodetic reference frame; and it is not validated for operational decisions such as landing-site certification or hazard clearance. Ice-prospectivity outputs regress a knowledge-driven fuzzy overlay rather than measured ice. The SomBench dataset card also notes heterogeneous spatial resolution and tens-of-meters absolute geolocation uncertainty. In short: this is a pretrained base for accelerating hypothesis generation and narrowing the search space — not a last-mile decision tool for mission operations.\n\n## Prithvi Series Extension: From Earth to the Moon, the \"Base Plus Data Plus Tooling\" Troika\n\nThe model joins the Prithvi family of open foundation models from IBM-NASA, previously covering geospatial, weather, and heliophysics, now extended to the Moon. NASA Chief Science Data Officer Kevin Murphy and IBM Research Europe UK\u002FIreland Director Juan Bernabe-Moreno both frame the release around \"starting from a shared base instead of building a new algorithm for every scientific question\" — a consistent narrative across the Prithvi series. The work was supported by NASA under Award No. 80MSFC25M0084. The full release is live on Hugging Face (nasa-ibm-ai4science\u002FNASA-IBM-Lunar-Foundation-Model) and the NASA-IMPACT GitHub repository.\n\n## So What: Why This Release Is Worth Its Own Write-Up\n\nThe Prithvi series extends the \"domain foundation model + publicly co-registered dataset\" pattern from Earth to the Moon. Laboratories working on lunar surface science no longer need to train a ViT from scratch or write glue code to align LROC, GRAIL, and Kaguya data across resolution tiers — they can fine-tune or LoRA-adapt the encoder to their specific task. This combination of \"dataset + pretrained base + downstream tooling, all open-source\" lands in the context of NASA's return-to-the-Moon program and effectively pushes lunar remote sensing from a closed-door effort by a few teams to something any Earth-based CV group can pick up. That is why IBM frames it as \"one of the first publicly available foundation models for scientific exploration of the Moon\" — the emphasis is not \"largest\" or \"strongest\" but \"public, modifiable, usable.\" The release also partially answers recurring scientific community questions about whether AI-for-Science tools can actually be open: weights on Apache-2.0, code on GitHub, and data on CC BY 4.0 all together is the furthest a comparable Earth-science model release has gone.\n\nReferences:\n- IBM newsroom official release (primary source): https:\u002F\u002Fnewsroom.ibm.com\u002F2026-09-10-ibm-and-nasa-release-open-source-ai-model-to-support-lunar-exploration\n- Unite.AI independent technical reading (full architecture, benchmark tables, training details): https:\u002F\u002Fwww.unite.ai\u002Fibm-and-nasa-open-source-lunar-foundation-model-with-sombench-dataset\u002F","nasa-ibm-lunar-foundation-model-sombench","2026-09-19T09:30:00Z","2026-09-19T09:06:17.353176Z","2026-09-19T09:06:17.353193Z",true,"agent",166,[43,51],{"slug":13,"tag_slug":13,"title_zh":44,"title_en":45,"intro_zh":46,"intro_en":47,"id":48,"is_active":39,"created_at":49,"modified_at":50},"AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":52,"tag_slug":52,"title_zh":53,"title_en":54,"intro_zh":55,"intro_en":56,"id":57,"is_active":39,"created_at":58,"modified_at":59},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":61},[62,67,72,77,82,87],{"id":63,"title":64,"news_slug":65,"published_at":66},"e406afb8-352e-4b01-947a-e63fbc7b072b","Ropedia 把 VLM 当规划器：S-Agent 用三级空间工具拼出 8B 空间智能体","ropedia-s-agent-spatial-3d-tool-mmsi","2026-06-21T08:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"94f00640-9d77-497c-906e-4018b1612f7f","Zyphra Zamba2-VL：Mamba2+Transformer 混血架构首次走进 VLM，TTFT 砍掉一个数量级","zyphra-zamba2-vl-mamba-transformer-ttft","2026-06-12T10:10:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"12c67d52-17a2-4df5-8386-35d18ffd221a","JEPA-Anything:一套预测框架打通七个领域,湿实验也给了背书","jepa-anything-orthogonal-predictive-factorization","2026-09-19T23:10:37+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"b9898ba4-65f5-4d80-b9ab-338b01fbd685","NASA 与 IBM 开源月球模型:极区找冰误差降 22%","nasa-ibm-lunar-foundation-model","2026-09-18T13:10:34+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"9ba1770e-87f2-47b9-aaa1-19f4f2ba78f1","ScienceIDE:把全球科学代码变成智能体训练场","scienceide-scientific-code-agent-environments","2026-09-17T23:05:17+00:00",{"id":88,"title":89,"news_slug":90,"published_at":91},"d056f67b-7e0d-4e44-8d39-e31ea50deeae","Bonsai 2 27B 三元压缩:Qwen3.8 压到 5.9 GB,benchmark 留存 98.2%","bonsai-2-27b-ternary-qwen3-8-compression","2026-09-17T15:47:00+00:00"]