#ArtificialIntelligence

A recent breakthrough study has discovered that the eyes, far from being mere passive players during sleep, actively orchestrate how our brains consolidate memories—guarding old knowledge while seamlessly incorporating new information. These findings, published by a Cornell University research team and highlighted in The Brighter Side of News on April 21, 2025, suggest that subtle changes in the pupil during sleep play a pivotal role in preventing memory “mix-ups” and could one day revolutionize how we learn, remember, and perhaps even treat memory disorders (The Brighter Side of News).

A groundbreaking international study led by over 150 scientists has produced the most detailed map ever of how visual information moves through the brain, uncovering more than 500 million intricate connections within a speck of mouse brain tissue and bringing the world closer to understanding how we see. Published in the journal Nature on April 9, 2025, the research combines genetic engineering, high-powered electron microscopy, and deep learning to capture not only the physical wiring of over 200,000 brain cells but also their real-time electrical activity in response to visual stimuli. The project—hailed as one of the most complex neuroscience experiments ever attempted—has generated a dataset of unprecedented size and detail: 1.6 petabytes, about the equivalent of 22 years of continuous high-definition video, all representing a single grain-sized fragment of brain.

A groundbreaking study from MIT is shaking up decades of neuroscience wisdom, revealing the brain’s “object recognition” pathway may also play a significant role in understanding spatial information—an insight that could revolutionize our approach to learning, artificial intelligence, and brain health around the world, including here in Thailand.

For years, scientists have believed the ventral visual stream, a key pathway in the human brain, is dedicated to recognizing objects—like a Starbucks cup on a Bangkok Skytrain or a rambutan vendor at the Chatuchak Market. This idea shaped not just neuroscience textbooks, but also inspired computer vision systems now used in everything from smartphones to smart cars. Yet, new research led by MIT graduate student Yudi Xie suggests the story is far more nuanced. Their findings, presented at the prestigious International Conference on Learning Representations, show that when deep learning models are trained not only to identify objects, but also to understand spatial features like location, rotation, and size, these models mirror neural activity in the ventral stream just as accurately as traditional object recognition models. In other words, the ventral stream might be wired for much more than recognizing faces or products—it could be a multifaceted toolkit for seeing and interacting with the world.