New Bayes-Markov V1 model maps brain's orientation selectivity with spiking neurons

A team of researchers—Abolfazl Moslemi, Milad Sarabadani, Fatemeh Sefidian, and Hossein Peyvandi—has published a computational re-implementation of Shirazi's Bayes-Markov model for orientation selectivity in the primary visual cortex (V1). The original model proposed that orientation-selective inhibition emerges from non-oriented LGN inputs through local probabilistic inference, using a maximum a posteriori (MAP) criterion over a two-layer hierarchical Markov random field. The new paper reconstructs this mathematical framework and implements a vectorized simulation that preserves the local clique operations while enabling systematic parameter sweeps.

The team evaluated the model with orientation tuning curves, an orientation selectivity index (OSI), controlled LGN noise perturbations, contrast tests, and comparisons across model variants. They also added a spiking SCI-layer implementation using leaky integrate-and-fire (LIF) and Hodgkin-Huxley neurons to test whether the rate-coded inhibitory field could be expressed through temporally explicit neural activity. Results confirm sharp orientation selectivity and robustness to moderate LGN noise, plus a biologically interpretable spiking realization of the inferred inhibitory field. This work bridges rate-based probabilistic models and spiking neural networks, offering a more biophysically grounded framework for understanding V1 computation and potentially guiding more efficient neuromorphic vision systems.