spike train encoding

To address the trade-off between biological plausibility and engineering practicality in spike-based encoders for neuromorphic computing, this paper proposes Spiketrum—a highly efficient, hardware-software co-designable spike encoder. Spiketrum is the first to enable lossless spike-domain compression of input data and exact, reversible reconstruction, while maintaining full compatibility with both spiking neural networks (SNNs) and artificial neural networks (ANNs). Through systematic cross-platform evaluation (FPGA + software), it outperforms existing biologically inspired encoders across multiple dimensions: classification accuracy, training speed, spike sparsity, entropy-based compression ratio, hardware resource utilization, and power consumption. By innovatively integrating spike coding theory, low-power digital circuit design, and a multi-classifier validation framework, Spiketrum significantly improves encoding efficiency and hardware energy efficiency. It provides a general-purpose encoding solution that simultaneously ensures biological interpretability and engineering deployability for neuromorphic systems.

Neuromorphic event cameras offer ultra-high temporal resolution but generate sparse binary spike streams, incurring substantial storage and transmission overhead. To address this, we propose the Spike Coding for Intelligence (SCI) paradigm—a joint compression-and-sensing framework. SCI features a bio-inspired dual-path network that separately models spatial semantics and motion dynamics; incorporates optical-flow-guided deformable alignment, motion-vector consistency refinement, and multi-dynamics temporal regression for robust spike sequence encoding. Crucially, SCI preserves downstream task performance while achieving high-fidelity, compact representations. Experiments demonstrate that SCI reduces BD-rate by 17.25% on average over state-of-the-art codecs, improves image classification accuracy by 4.3% relative to SpiReco, lowers encoder computational complexity by 88.26%, and cuts inference latency by 42.41%.

This work addresses the limitations of traditional spike coding methods, which often rely on probabilistic models and lack compatibility with mainstream signal processing theory, making it difficult to define bandwidth and guarantee reconstruction fidelity. The authors propose a novel spike coding framework grounded in causal temporal wavelets, introducing for the first time a bandwidth-controllable wavelet representation into spike coding to achieve sparse, localized, and reconstructable spiking representations of temporal signals. By integrating causal bandpass wavelet frames, spike-based quantization, and temporal discretization, the method provides rigorous theoretical bounds on reconstruction error. Experimental results demonstrate that, on ECG and audio signals, the proposed approach achieves normalized root-mean-square errors comparable to those of the continuous wavelet transform while remaining amenable to deployment on neuromorphic hardware.

Spiking neural networks (SNNs) suffer from limited performance on visual tasks, primarily due to inadequate information representation in conventional spike encoding schemes and poor compatibility with surrogate gradient-based training. To address this, we propose the first hybrid time-bit spike encoding framework designed explicitly for end-to-end surrogate gradient training. Our method decomposes input images into bit planes and applies temporal coding—such as latency or phase coding—to each bit plane, yielding a joint rate-time representation. Crucially, the entire encoding process is differentiable, enabling stable and efficient backpropagation via surrogate gradients. This design substantially enhances spatiotemporal information capacity and gradient propagation stability. Extensive experiments on CIFAR-10/100 and ImageNet-1K demonstrate that SNNs trained with our framework match or surpass state-of-the-art encoding methods—including rate coding, time-to-first-spike (TTFS), and spike-time coding—in accuracy, validating its effectiveness, generalizability, and seamless integration with standard surrogate-gradient optimization pipelines.

This work proposes MetaSort, a novel algorithm that unifies neural spike compression and classification within a single framework—addressing a key limitation of existing approaches that treat these tasks separately and thus struggle to balance efficiency and performance. MetaSort achieves high-fidelity non-uniform compression through adaptive level-crossing sampling and constructs a geometry-aware latent representation by integrating meta-transfer learning with the intrinsic geometric structure of the data, enabling effective few-shot classification. Evaluated on in vivo neural spike recordings, the method significantly improves both compression ratio and classification accuracy, offering an efficient and practical solution for ultra-low-power on-chip neural signal processing.

This work addresses the inherent mismatch between continuous speech signals and the discrete, event-driven nature of spiking neural networks (SNNs), where conventional fixed encoders struggle to produce task-optimized spike representations. The authors propose a learnable residual speech-to-spike encoder trained end-to-end with a recurrent leaky integrate-and-fire (R-LIF) SNN, directly optimizing for class separability rather than signal reconstruction. This approach achieves the first parameter-efficient, adaptive speech encoding for SNNs and provides a systematic evaluation of biologically plausible learning rules—such as Direct Feedback Alignment (DFA)—on audio tasks. On Google Speech Commands v2 (GSC-v2), the method attains 94.97% accuracy; a compact variant with only 35k parameters reaches 89.8%; and DFA achieves 91.5%, demonstrating its practical viability.

Directly training spiking neural networks (SNNs) on static images leads to temporal collapse and hinders effective modeling of spatiotemporal dynamics, as conventional time encoding schemes—e.g., repeated frame presentation—induce rate-coding bias rather than exploiting rich temporal structure. Method: We propose a learnable phase-shift temporal encoding mechanism that maps static images to spike trains with adaptive timing patterns. Crucially, we decouple encoding design from network optimization, revealing that convolutional layer learnability and surrogate gradient formulation—not the encoding itself—are the primary determinants of performance. Accordingly, we design a minimal, end-to-end trainable temporal encoder. Results: Our approach significantly narrows the accuracy gap between direct and rate coding, preserves the energy efficiency of SNNs, and enhances spatiotemporal feature representation. It establishes a new paradigm for efficient, temporally expressive modeling of static images in SNNs.

本文通过使用非学习型高阶可编程编码器将音频转化为尖峰信号，并在FPGA上实现，优化了编码器和分类器以提高效率并降低能耗。

It remains unclear whether the information content of individual spikes in population neural coding models—without spike sorting—depends on spike amplitude and prior spiking history. Method: Using multi-neuron marked point process data from rat hippocampus, we quantified the information carried by isolated spikes versus conditionally decoded spikes (i.e., incorporating preceding spike sequences), employing information-theoretic entropy measures, state-space modeling, and a clusterless decoding framework. Contribution/Results: While low-amplitude spikes convey significantly less information than high-amplitude ones when analyzed in isolation, their conditional information—when contextualized by recent spiking history—approaches or equals that of high-amplitude spikes. This is the first systematic demonstration that poorly clusterable spikes are not noise but encode critical temporal information. The findings provide theoretical justification and empirical support for abandoning traditional spike sorting and advancing continuous-feature-based neural decoding paradigms.

为解决低比特脉冲神经网络的计算和存储成本问题，提出BASC框架，通过时间行为校正和边界级通道校正方法优化量化和剪枝。