Miguel de Prado

A biologically grounded framework in which noisy neural and synaptic dynamics perform inference and learning via stochastic sampling from an internal energy function, capturing uncertainty over latent states and model parameters through neural and synaptic variability, respectively is discussed.

This report demonstrates using a detailed transaction-level model (TLM) of a probabilistic analogue in-memory computing (AIMC) processor that the same energy-based model can execute well over 1000x faster than data-center-grade hardware by eliminating the HBM interface and performing computation directly within on-chip memory.

Achieving up to 5x speedups in Bayesian inference on edge devices could revolutionize real-time decision-making in resource-limited settings.

You can now train Gaussian Splatting models on your edge device, thanks to a clever optimization that slashes memory use by 8x and speeds up training by 4x, all without sacrificing reconstruction quality.