Google Public Sector Fuels Caltech AI Model Efficiency

Modern research workloads, driven by AI and HPC, demand processing of structured and unstructured data at an unprecedented scale, maintaining sub-millisecond storage latency, enterprise-level security, compliance, and reproducibility. Caltech has historically led large-scale AI research.

Google's support provides Cloud GPUs, Google's custom Arm-based Axion processors, and Cloud Tensor Processing Units (TPUs) for intense workloads. Researchers gain access to Google's first-party datasets, including AlphaFold (from DeepMind), Earth Engine, Google Maps Platform, and VirusTotal.

The initiative also includes the fully-managed Vertex AI Platform, featuring Vertex AI Agent Builder and over 200 first-party (Gemini, Imagen 3), third-party, and open (Gemma) foundation models in Model Garden. Dedicated campus training and workshops will increase AI literacy and adoption.

This infrastructure integrates with Caltech’s existing HPC environments. Dr.

Babak Hassibi, Mose and Lillian S. Bohn Professor at Caltech, will lead an initiative using Vertex AI to develop training methods incorporating pruning, quantization, and distillation. This work aims to significantly reduce inference time costs of trained models, making AI more accessible and sustainable, and improving system safety for AI at the edge.

Reymund Dumlao, Director of State & Local Government and Education at Google Public Sector, stated this support enables scientific discoveries across all domains.