Neural encoding of competing hypotheses and computational inference

• Neural activity patterns can encode competing hypotheses about which option will lead to the correct outcome (e.g., which landmark leads to the correct destination).¹
• The brain distinguishes between such ambiguous hypotheses via neural encoding and computational inference.^{1 1}

Gardner updates

• Neural activity patterns can encode competing hypotheses about which landmark will lead to the correct destination. ¹

Weekly enrichment (2026-07-20)

• The finding comes from a mouse study of the retrosplenial cortex (RSC) and is described as the first time neural activity patterns encoding simultaneous competing hypotheses have been observed in the brain.²
• Mice foraged in a round arena with 16 wall ports; the reward sat at the port by the counterclockwise of two identical light dots that were only visible up close, forcing animals to hold multiple location hypotheses until disambiguation.²
• The researchers recorded 50–90 simultaneous layer-5 RSC neurons in four mice using tetrode array drives alongside behavioral tracking.³
• RSC populations showed distinct activity patterns for each hypothesis that collapsed into the single correct pattern once the mouse got close enough to identify the dots.²
• Hypotheses were encoded as separate locations in neural activity space, with divergent trajectories for identical sensory inputs, enabling correct interpretation without external context cues at the moment of disambiguation.³
• Both RSC and artificial recurrent neural networks encoded mixtures of hypothesis, location, and sensory information constrained by robust low-dimensional dynamics; interconnectivity appeared key to holding two hypotheses at once.³
• The paper (Voigts, Kanitscheider, Miller, Toloza, Newman, Fiete, Harnett) appeared in Nature Neuroscience on June 6, 2025 (28(6):1293–1299), from MIT’s McGovern Institute and HHMI Janelia, funded by NIH, the Simons Center for the Social Brain, NIGMS, and NSF’s Center for Brains, Minds, and Machines.^{2 3}
• Decoding/BCI implication: explicit, low-dimensional population codes for competing hypotheses offer a substrate that neural decoders could read out for sequential reasoning and belief-state tracking, beyond static stimulus decoding.³

Footnotes

1. https://news.mit.edu/2025/how-brain-distinguishes-between-ambiguous-hypotheses-0606 ↩ ↩² ↩³ ↩⁴

2. https://news.mit.edu/2025/how-brain-distinguishes-between-ambiguous-hypotheses-0606 ↩ ↩² ↩³ ↩⁴

3. https://doi.org/10.1038/s41593-025-01944-z ↩ ↩² ↩³ ↩⁴ ↩⁵