This study trained a neural network using deep reinforcement learning to exhibit primate-like perceptual decision-making abilities. The network learned key characteristics of primate decision-making, such as trading off speed for accuracy and flexibly changing its mind in response to new information, on a noisy perceptual discrimination task.

The internal dynamics of these networks suggest that these abilities are supported by decision mechanisms similar to those observed in primate neurophysiological studies. These results provide experimental support for key pressures that led to the emergence of primate flexible decision-making capabilities.

👉 Read the full article on arXiv

This paper proposes a new perspective termed "polyphonic intelligence," contrasting with dominant models that prioritize convergence by reducing uncertainty and eliminating competing explanations. It focuses on biological and adaptive systems that maintain redundancy and ambiguity.

Polyphonic intelligence posits that coherent behavior and meaning emerge from the coordination of multiple semi-independent inferential processes operating under shared constraints. A variational framework is introduced where multiple coordinated approximations are maintained without winner-takes-all selection, demonstrating how plurality can remain stable, tractable, and productive. Proof-of-principle examples show that non-dominating, pluralistic inference can be implemented in simple computational systems without requiring centralized control or global convergence.

👉 Read the full article on arXiv

This study addresses the need for accurate, low-latency estimates of instantaneous phase for oscillations, crucial for closed-loop sensing and actuation. The endpoint-corrected Hilbert transform (ecHT) improves phase estimation at the most recent sample by applying a causal narrow-band filter to the analytic spectrum, reducing boundary artifacts of the Hilbert transform.

The research analytically derives the ecHT endpoint operator, demonstrating its output decomposes into a desired positive-frequency term and a residual leakage term that sets an irreducible variance floor. This leads to an explicit characterization and bounds for endpoint phase/amplitude error, a mean-squared-error-optimal scalar calibration (c-ecHT), and practical design rules relating window length, bandwidth/order, and center-frequency mismatch to residual bias via an endpoint group delay. The resulting calibrated ecHT achieves near-zero mean phase error and remains computationally compatible with real-time pipelines.

👉 Read the full article on arXiv