PulseAugur
EN
LIVE 19:01:47

TESSERA v2 study reveals optimal scaling for Earth-observation models

By PulseAugur Editorial · [2 sources] ·

Researchers have conducted a large-scale study on scaling pixel-wise Earth-observation foundation models, involving 395 training runs on 1,024 NVIDIA GH200 superchips. The study found that pretraining loss is a poor predictor of downstream performance, suggesting that focusing compute on larger encoders and more data, rather than larger projectors, is more effective. The team trained a family of models, including the 21-million-parameter TESSERA v2-1B-M, which outperforms larger open and proprietary models and offers efficient serving through Matryoshka representations. AI

IMPACT Provides a data-driven recipe for efficiently scaling Earth-observation foundation models, potentially accelerating deployment and improving performance.

RANK_REASON The cluster describes a research paper detailing a large-scale empirical study on scaling foundation models for Earth observation.

Read on Hugging Face Daily Papers →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

TESSERA v2 study reveals optimal scaling for Earth-observation models

COVERAGE [2]

  1. arXiv cs.LG TIER_1 English(EN) · Zhengpeng Feng, Sadiq Jaffer, Ira Shokar, Jovana Knezevic, Mark Elvers, Clement Atzberger, Robin Young, Aneesh Naik, Niall Robinson, Andrew Blake, David Coomes, Anil Madhavapeddy, Srinivasan Keshav ·

    TESSERA v2: Scaling Pixel-wise Earth Foundation Models

    arXiv:2607.03949v1 Announce Type: cross Abstract: Pixel-wise Earth-observation (EO) foundation models are now achieving state-of-the-art performance via generated spatial embeddings. However, how these models scale and how best to spend a pretraining budget remain poorly understo…

  2. Hugging Face Daily Papers TIER_1 English(EN) ·

    TESSERA v2: Scaling Pixel-wise Earth Foundation Models

    Large-scale controlled experiments reveal optimal scaling strategies for Earth-observation foundation models, enabling efficient training and deployment through encoder growth, downstream performance selection, and model distillation.

RELATED ENTITIES

RELATED TOPICS