# ML Times

## Breaking the Memory Barrier: Near Infinite Batch Size Scaling for Contrastive Loss

- **Near infinite batch size scaling** for contrastive loss is achieved through a novel tile-based computation strategy that avoids the full instantiation of the similarity matrix, significantly enhancing performance in representation learning.

## Detecting when LLMs are uncertain

- **Entropix** introduces **adaptive sampling** techniques to enhance LLM reasoning during uncertainty, aiming to improve decision-making by analyzing token distributions.

## Universal optimality of Dijkstra via beyond-worst-case heaps

- This paper establishes that **Dijkstra's algorithm** achieves **universal optimality** in both running time and comparisons when paired with an efficient heap, marking a significant advancement in graph algorithm performance guarantees.

## OmniParser for Pure Vision Based GUI Agent

- **OMNIPARSER** enhances the capabilities of **GPT-4V** by enabling it to accurately parse user interface screenshots, identifying interactable icons and understanding their semantics for improved action generation.

## The open secret of open washing – why companies pretend to be open source

- **Open washing** is a deceptive practice where companies falsely claim their products are open source, undermining the true spirit of transparency and collaboration, as seen with Meta's Llama 3, which fails to meet the [Open Source Definition](https://opensource.org/osd).

## ML accelerated with TEE + Federated Learning

- **ML accelerated** through **TEE** (Trusted Execution Environment) and **Federated Learning** offers a promising solution for secure data sharing in oncology, described as "plug and play" by the research lead.

## Intel GPU Support Now Available in PyTorch 2.5

- **Intel GPU support in PyTorch 2.5** enhances performance and functionality for Intel® Arc™ graphics and Intel® Data Center GPU Max Series, integrating the SYCL software stack for a seamless user experience in AI applications.