Inspired by the brain's ability to focus only on what matters, a new AI model processes videos 65% faster and more accurately by ignoring most visual frames.
The Research
Researchers at the Japan Advanced Institute of Science and Technology (JAIST), led by Professor Shogo Okada and doctoral student Hung Le, developed EMF-dVAE (Efficient Multimodal Fusion with a discrete Variational Autoencoder). The model mimics how humans selectively attend to important moments in a conversation, using audio cues to decide when to analyze visual information.
During training, the AI learns which visual features are most informative by reconstructing corrupted video data. At inference, it only processes the key visual segments, combining them with audio and language data for final predictions.
On the ETS-Interview dataset of 1,891 two-minute job interview videos, EMF-dVAE achieved state-of-the-art accuracy while using only 15.42% of visual features. It reduced processing time per video from 52 seconds to 18 seconds—a 65% reduction—while also improving accuracy because redundant frames often add noise.
The study was published in the journal Information Fusion (volume 137, January 2027).
Why It Matters
This brain-inspired approach highlights how selective attention—a key cognitive ability—can dramatically boost efficiency. For everyday tasks, filtering out irrelevant information can help you think faster and make better decisions, whether you're studying, working, or just trying to remember what really matters.
What You Can Do
Train your selective attention by practicing mindfulness or playing attention-focused games. Before starting a task, explicitly decide what to ignore. This reduces mental clutter and improves performance, just as EMF-dVAE improves AI.
Source: Neuroscience News
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
