Summary: A new study reveals how the human brain efficiently encodes and retrieves visual memories by organizing them into categories and relying on the precise timing of neuronal spikes. Researchers recorded hippocampal activity from epilepsy patients as they viewed and later recalled images from five object categories, then applied machine learning to decode which category the subject was remembering based solely on neural timing patterns.
The findings indicate that the hippocampus does not store every object as an independent memory trace. Instead, it compresses visual information by grouping objects into categories and using millisecond-scale spike timing to represent those categories. This insight advances our understanding of episodic memory and suggests paths toward brain-computer interfaces and memory prostheses for people with memory impairments.
Key Facts:
- Category-based Storage: Visual objects are encoded as categories rather than as unique, individual items, reducing memory complexity.
- Temporal Coding: The timing of neural spikes—when neurons fire at the millisecond level—carries critical information about which memory category is active.
- Clinical Potential: These discoveries can inform the development of memory prostheses and other neurorestorative tools for conditions that affect memory, including dementia and hippocampal dysfunction.
Source: USC
USC researchers have clarified how the hippocampus represents visual memories.
Published in Advanced Science, the study combines intracranial human recordings with an interpretable machine learning decoding model to map how hippocampal neurons represent categories of visual memories. Think of it as revealing the brain’s internal filing system for visual information: rather than storing countless individual images, the hippocampus appears to file them into a limited set of meaningful categories.
Using recordings of spike timing from hippocampal neurons, the team was able to determine, with strong reliability, which category of image a participant was recalling. The results answer a longstanding question about whether hippocampal memory representations are object-specific or category-based, while showing that temporal patterns of spikes—rather than average firing rates alone—carry category information.
The research was led by Dong Song, associate professor in the Department of Neurological Surgery and the Alfred E. Mann Department of Biomedical Engineering, and Charles Liu, director of the USC Neurorestoration Center at Keck School of Medicine and professor of biomedical engineering at USC Viterbi. The first author, Xiwei She, is a former Ph.D. student in Song’s lab and now a postdoctoral researcher at Stanford University.
How does the brain store visual information?
The hippocampus is essential for forming episodic memories—the what, where, and when of past experiences. While its role in encoding spatial and temporal aspects of memory is well established, encoding the vast diversity of objects (the “what”) poses a major challenge given the high-dimensional nature of visual input. Storing every object individually would be inefficient; categorization offers a practical solution.
Song’s lab has been developing memory prosthesis technology aimed at restoring cognitive function in clinical populations. This study was designed not only to support those translational goals but also to resolve fundamental questions about hippocampal coding strategies.
Human intracranial recordings provide direct insight
The team recorded spikes from hippocampal CA3 and CA1 neurons in 24 epilepsy patients who had depth electrodes implanted for clinical monitoring. These direct human recordings enabled analysis of precise spike timing during a delayed match-to-sample (DMS) task, a standard test of visual short-term memory.
Participants viewed images from five categories—animal, plant, building, vehicle, and small tools—and later attempted to recall or recognize those images. The researchers applied an interpretable decoding model to the spike trains to ask whether the category of a recalled image could be inferred from hippocampal activity alone.
The decoding model successfully predicted the memory category from temporal spike patterns, demonstrating that the hippocampus represents visual memories in a category-specific way and that precise timing is a key information carrier.
An efficient, distributed coding strategy
Rather than relying on single, highly selective neurons, the hippocampus appears to use a distributed population code: ensembles of neurons collectively represent memory categories, while individual neurons contribute through brief, temporally specific moments of activity. The study found that most neurons (approximately 70–80%) participated in category assignment, though each neuron’s contribution was localized in time. This approach balances representational capacity with metabolic efficiency.
CA3 and CA1 neurons carried similar and partially redundant category information, consistent with strong feedforward connections between these hippocampal subregions. Identifying the optimal temporal resolution for decoding each category in each neuron highlighted the importance of millisecond-scale timing for memory coding.
These discoveries pave the way for improved neural decoding algorithms and could inform the design of clinical devices aimed at restoring or augmenting human memory.
About this visual memory research news
Author: Amy Blumenthal
Source: USC
Contact: Amy Blumenthal – USC
Image: Credit to Neuroscience News
Original Research: Open access. “Distributed Temporal Coding of Visual Memory Categories in Human Hippocampal Neurons Revealed by an Interpretable Decoding Model” by Dong Song et al. Published in Advanced Science.
Abstract
Distributed Temporal Coding of Visual Memory Categories in Human Hippocampal Neurons Revealed by an Interpretable Decoding Model
The hippocampus is essential for forming episodic memories. While its role in encoding spatial and temporal information is well documented, how it represents objects—the “what” of memory—remains unclear due to the high dimensionality of object space. To reduce complexity, the hippocampus may encode categories rather than individual items.
This study uses an experimental-modeling approach to investigate visual memory category coding in humans. Spikes were recorded from CA3 and CA1 hippocampal neurons in 24 epilepsy patients performing a delayed match-to-sample task with five image categories. An interpretable decoding model was applied to hippocampal spiking activity to decode memory categories and reveal spatio-temporal characteristics of encoding.
The model estimated optimal temporal resolutions for decoding each visual category at the single-neuron level. Results show that visual memory categories can be decoded from hippocampal spike patterns, supporting category-specific coding. Ensembles of hippocampal neurons encode categories in a distributed population code, while individual neurons employ a temporal code. CA3 and CA1 neurons provide similar, partly redundant category information, consistent with strong feedforward synaptic connections between these regions.