The field of neuromorphic computing and spiking neural networks is moving towards more biologically plausible and energy-efficient models. Researchers are exploring alternatives to traditional backpropagation algorithms, such as forward-only approaches and parameter-shift rules, to enable on-device learning and neuromorphic computing. Another area of focus is the development of scalable and efficient interconnects for large-scale neuromorphic architectures, as well as the creation of open-source frameworks for emulating and testing spiking neural networks. Furthermore, innovative learning rules and methods are being proposed to improve the performance and flexibility of spiking neural networks, including the extension of spike-timing dependent plasticity to learn synaptic delays. Noteworthy papers include: Forward Target Propagation, which proposes a forward-only approach to global error credit assignment via local losses. NeuroCoreX, an open-source FPGA-based spiking neural network emulator with on-chip learning, provides a flexible co-design and testing platform for SNNs.
Advances in Neuromorphic Computing and Spiking Neural Networks
Sources
Forward Target Propagation: A Forward-Only Approach to Global Error Credit Assignment via Local Losses
Mapping and Scheduling Spiking Neural Networks On Segmented Ladder Bus Architectures
Gradients of unitary optical neural networks using parameter-shift rule
Feedforward Ordering in Neural Connectomes via Feedback Arc Minimization
NeuroCoreX: An Open-Source FPGA-Based Spiking Neural Network Emulator with On-Chip Learning
A Scalable Hybrid Training Approach for Recurrent Spiking Neural Networks
Extending Spike-Timing Dependent Plasticity to Learning Synaptic Delays