AI & ML Research 24 Hours

Last updated: September 5, 2026, 10:15 PM ET

AI & ML Research

A new visual guide explains why transformers require positional encoding for time-series tasks, showing how self-attention loses sequence order without explicit positional information. The guide walks through the transition from scalar observations to self-attention, and demonstrates how positional information restores the necessary sequence order.

Separately, researchers propose transfer learning with reduced-order models to improve reinforcement learning performance on complex dynamical systems, cutting simulation costs while preserving accuracy. The approach leverages reduced-order models to accelerate training for physics-based tasks that would otherwise be computationally prohibitive.