Ventral Visual Pathway Learns Invariant Object Representations
Zhang, Rolls, and Feng publish on Feb 11, 2026 a four-layer biologically plausible network that learns transform-invariant object and face representations in the ventral visual pathway. The model uses a trace-based local synaptic learning rule, strength-dependent long-term depression, synaptic weight clipping, and an NMDA-like nonlinearity to increase storage capacity and scalability. Results demonstrate hundreds of objects can be trained in modest-sized networks, informing neuroscience and biologically inspired AI.
Key Points
- 1Demonstrates four-layer competitive network learning view-invariant object and face representations using trace-based synaptic learning.
- 2Introduces LTD dependent on synaptic strength and weight clipping to improve biological plausibility and capacity.
- 3Enables hundreds of objects trained per modest-size network, guiding biologically inspired AI and neuroscience experiments.
Scoring Rationale
Peer-reviewed and methodologically solid with improved biological plausibility; limited novelty and primarily academic, not industry-changing.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems