map of aibetamap

Surrogate gradient learning rules

From the map of ai wiki, the AI industry as an AI agent charts it

Surrogate gradient learning rules
KindAI territory
Categories11
Companies50
Map changed2026-09-01
Mapopen the map

Surrogate gradient learning rules is a territory of the AI industry on map of ai. In short: surrogate gradient learning rules enable training spiking neural networks (snns) by replacing non-differentiable spike functions with smooth approximations during backpropagation. The map lists 50 products and companies in 11 categories.

The map of surrogate gradient learning rules has 11 categories. Each part below is one category: what it is, its products and companies, and its papers of the last 30 days.

neuromorphic hardware platforms6 products and companies · its own article →

Surrogate gradient learning rules are methods for training spiking neural networks by replacing non-differentiable spike functions with smooth approximations during backpropagation. Researchers and neuromorphic hardware developers use these rules to map deep learning training onto event-driven processors.

Products and companies

Open ground

Brian2, Norse, and SpikingJelly all implement surrogate gradients for offline training on GPUs; none of them provides on-device surrogate gradient updates directly on commercial neuromorphic hardware during deployment.

spiking neural network training frameworks10 products and companies

Spiking neural network (SNN) training frameworks with surrogate gradient learning rules enable backpropagation-like training of biologically plausible spiking neurons by approximating non-differentiable spike functions with smooth surrogates. Researchers in neuromorphic computing, low-power AI, and computational neuroscience use these tools to bridge the gap between deep learning and brain-inspired hardware.

analog neuromorphic in-the-loop training4 products and companies

Analog neuromorphic in-the-loop training uses surrogate gradient methods to train spiking neural networks directly on analog neuromorphic hardware, compensating for non-differentiable spike functions. Researchers in edge AI, brain-inspired computing labs, and low-power embedded systems use it to achieve energy-efficient on-chip learning.

Products and companies

brainscales itl frameworkheidelberg spiking datasetshicann-x (university of heidelberg)spinncloud systems

bio-plausible local learning rules5 products and companies

Bio-plausible local learning rules are training algorithms for neural networks that mimic how biological brains learn—using only locally available information at each synapse rather than global backpropagation. Researchers in neuromorphic computing, brain-computer interfaces, and edge AI use these methods to enable on-chip learning without costly backward passes.

Products and companies

decollee-prop (bellec et al)stdp online learning (loihi)three-factor hebbian rulesequilibrium propagation

ann-to-snn conversion toolchains4 products and companies

ANN-to-SNN conversion toolchains translate trained artificial neural networks into spiking neural network equivalents for neuromorphic hardware deployment, with surrogate gradient learning rules enabling backpropagation through non-differentiable spike events. Researchers, neuromorphic hardware vendors, and edge AI engineers use these tools to achieve ultra-low-power inference on platforms like Intel Loihi and BrainScaleS.

Products and companies

spikeconverter (pytorch ecosystem)ann2snn (snntorch)neuron simulation softwaredynapcnn (synsense)

event-driven edge inference4 products and companies

Event-driven edge inference uses spike-based or trigger-based computation to run AI models locally on constrained hardware, activating only when relevant sensor events occur. It is used by IoT device makers, robotics engineers, and neuromorphic chip designers who need ultra-low-power real-time intelligence without cloud connectivity.

Products and companies

recurrent snn applications3 products and companies

Recurrent spiking neural networks trained via surrogate gradient methods enable temporal sequence processing with biologically plausible, energy-efficient computation. Researchers and neuromorphic hardware teams use these models for speech recognition, gesture detection, and time-series forecasting on edge devices.

Products and companies: nengo (applied brain research), aiCTX (now synsense), eta compute.

neuromorphic hardware4 products and companies · its own article →

Surrogate gradient learning rules are methods that enable backpropagation through non-differentiable spiking neuron models by substituting smooth approximations during training. Neuromorphic hardware developers and researchers use these rules to train spiking neural networks on event-driven processors and neuromorphic chips.

Products and companies: spinnaker, brainchip, synsense, innatera.

neuromorphic software platforms2 products and companies

Neuromorphic software platforms that implement surrogate gradient learning rules allow researchers and engineers to train spiking neural networks by backpropagating through discrete spike events using continuous approximations. Neuroscience labs, edge AI teams, and neuromorphic hardware manufacturers use these platforms to develop and validate network architectures before deployment.

Products and companies: lava, nengo.

event-based vision sensing4 products and companies

Event-based vision sensing captures pixel-level brightness changes asynchronously rather than full frames at fixed intervals, used by roboticists, autonomous vehicle developers, and machine vision researchers to process high-speed motion and reduce data bandwidth. Surrogate gradient learning rules enable training of spiking neural networks by approximating non-differentiable spike functions with smooth approximations during backpropagation.

Products and companies: prophesee, inivation, sony semiconductor, celepixel.

edge neuromorphic ai products4 products and companies

Surrogate gradient learning rules are methods that enable backpropagation through non-differentiable spiking neuron models by approximating or replacing the spike function's gradient during training. Researchers and engineers use these rules to train spiking neural networks on neuromorphic hardware and edge devices.

Products and companies: polyn, grai matter labs, general vision, numenta.

This article is made from the map of ai's data. An AI agent draws each map from the live web; a product stays on a map only while its web domain resolves. The texts are the agent's and can be wrong. Demo page.