Latest News

The SustainML Framework has extended its metadata management capabilities by integrating a new module for automated dataset profiling and semantic description generation using Large Language Models (LLMs). This development enables automatic generation of dataset metadata — including descriptions, semantic classifications, and contextual insights — directly from raw tabular or structured data.

Researchers from the German Research Center for Artificial Intelligence (DFKI) have released a timely and critical study on how irregular sampling in time-series data affects AI performance in Human Activity Recognition (HAR). Presented at the International Conference on Pervasive Computing and Communications (PerCom), the paper "Assessing the Impact of Sampling Irregularity in Time Series Data: Human Activity Recognition as a Case Study" exposes limitations in both discrete-time and continuous-time neural networks when faced with real-world sensor noise, packet loss, and jitter.

IBM Research has published a landmark study in Neuron (2024) that bridges neuroscience and sustainable AI, quantifying the energy-performance tradeoffs of brain-inspired computing. The paper, “The Brain, Energy, and AI”, offers a first-of-its-kind analytical framework for comparing biological and artificial neural systems under a common lens of energy efficiency and information processing.

SustainML integrates Retrieval-Augmented Generation (RAG) to intelligently recommend machine learning models to users — without retraining or fine-tuning large language models (LLMs). Instead, SustainML combines vector search, graph-based reasoning, and language generation to query a massive repository of models hosted on Hugging Face.