A von-Neumann inspired hypothesis of efficient computation as an alternative for neural decoding, a memory-augmented Transformer whose feed-forward block is a low-rank instruction bank: a shared base operator plus a small set of learned low-rank instructions, from which a per-token code synthesizes the weight matrix actually used at that token.
Abstract
Cortical computation is strikingly low-dimensional: a handful of latent variables, carried in a neural population's activity, steer the higher-dimensional responses of individual neurons. Our aim is sample efficiency-models that decode well from limited data and at small parameter budgets. In a standard Transformer layer, the feed-forward block applies the same operator to every token. We suggest a von-Neumann inspired hypothesis of efficient computation as an alternative for neural decoding: a controller decodes an instruction and then executes a token-specific operator; the usual realization-a soft mixture of experts-only blends their outputs, not operators. We introduce a von-Neumann State-Space Transformer (VN-SST), a memory-augmented Transformer whose feed-forward block is a low-rank instruction bank: a shared base operator plus a small set of learned low-rank instructions, from which a per-token code synthesizes the weight matrix actually used at that token. The code is read from a low- dimensional projection of a carried state-space memory, so a slow latent trajectory acts as an instruction pointer-mirroring how low-dimensional dynamics may route cortical computation. On three motor-cortex neural-decoding benchmarks, VN-SST is far more data-efficient than a modern Transformer, each jointly predicting spikes and decoding behavior. This model wins by a wide margin on the scarcest benchmark, leads on the other two, and turns longer context into rising rather than falling accuracy. We evaluated that the network compresses a large instruction bank to a few bits per token, so program capacity acts as a control channel, not an accuracy lever. The same model is also more parameter-efficient on two small text benchmarks used for language modeling (LLMs), suggesting a generic mechanism.
This study proposes a low-complexity, bidirectional, single-pass Elman recurrent neural network detector for binary phase-shift keying signals transmitted with faster-than-Nyquist signaling. Since the faster-than-Nyquist intersymbol interference has a short, finite memory, the classical Elman recurrent neural network, which contains no gating mechanism, is a sufficient and parameter-efficient model. The proposed detector processes the received sequence in both forward and backward directions in a single pass, forming a learned counterpart of the optimal BCJR forward-backward recursion. Under a root-raised-cosine pulse over an additive white Gaussian noise channel, simulations for two compression factors show that the proposed detector, with only twenty-five to sixty-five trainable parameters, attains a bit error rate very close to that of the M-BCJR algorithm, while reducing the look-up-table hardware cost by thirty-eight to forty-six percent and using no explicit division or exponential operations. A more compact configuration offers up to a sixty-seven percent reduction at a small performance penalty. To the best of our knowledge, this is the first study to investigate the classical Elman recurrent neural network architecture for the faster-than-Nyquist detection problem.
Nurettin Safak, Osman Tokluoğlu, E. Çavuş· 0 citations
We introduce CMP (Cognitive Memory Primitive), an architecture that represents inputs as sparse relational codes, stores them in a two-tier competitive memory, and learns entirely through local, gradient-free updates, with no backpropagation anywhere in the network. We use this architecture to test a specific hypothesis: that catastrophic forgetting, usually treated as a training-time defect to be patched with replay or regularization, is instead a structural consequence of how backpropagation assigns credit and that a learning rule that is local and sparse by construction should resist it without a patch. On a controlled domain-incremental protocol across 15 text domains, three-seed replicated, CMP's backward transfer is 15-19x better than a matched-size Transformer trained with online EWC, and the result survives a domain-order control (reported as a range, +0.24 to +0.44, rather than a single figure). We report this alongside a real, substantial accuracy gap versus the Transformer baseline, a null result on a recognized vision benchmark, and a diagnosed, unresolved failure attempting to combine this architecture with a separate mechanism that improves raw accuracy, disclosed because an honest negative result is more useful than an omitted one. The central claim is narrow and falsifiable: local, sparse, non-backpropagation learning measurably resists catastrophic forgetting better than backpropagation with its standard fix, under conditions we state precisely.
Feedforward network (FFN) blocks account for a large fraction of the parameters and computation in Transformer architectures, yet their internal structure remains difficult to interpret due to the additive superposition induced by the residual stream. We examine whether the activation of an FFN neuron can be explained by a sparse set of preceding neuron activations and attention outputs. We introduce a training-free attribution method that estimates the relative influence of upstream neurons and attention outputs on a target neuron's activation. Empirically, across models and layers, we find that small subsets of preceding activations and attention outputs suffice to preserve neuron activations with high fidelity when all remaining inputs are masked with their average values. Effective sparsity is even greater when accounting for the inherent activation sparsity of upstream layers. Moreover, applying the neuron-specific masks in all layers simultaneously, such that the induced deviations propagate through the network, leaves model perplexity largely unchanged at moderate sparsity levels. These results demonstrate that, despite dense parameterization, FFNs exhibit sparse and structured inter-layer dependencies at the neuron level. Our method provides a practical, scalable tool for circuit-level interpretability and identifies candidate sparse pathways with potential implications for efficient inference.
Extensive experiments on static and neuromorphic benchmarks show that lower-bit BASC models match or outperform higher-bit baselines and retain this accuracy advantage after structured pruning, while further reducing model storage and synaptic operations.
Linliang Chen, Yan Zhong, Xin Liu et al.· 0 citations
Hard-constrained recurrent physics-informed networks (HRPINNs) embed known dynamics inside a recurrent numerical integrator and restrict a neural branch to learning only the residual dynamics that the first-principles model does not capture. Kolmogorov--Arnold Networks (KANs) have been proposed as parameter-efficient replacements for multilayer perceptrons (MLPs) in such residual branches, but their learnable B-spline activations follow a markedly different execution profile. Building on prior work that characterized when a vanilla B-spline KAN matches or underperforms an MLP as an HRPINN residual branch in discovery accuracy, this paper asks whether that parameter efficiency survives deployment. Using identical trained weights, we measured execution latency, energy per integration step, and dependability under post-training quantization in the closed recurrent loop on a RISC-V RV64GC platform without vector extensions (StarFive VisionFive~2, SiFive U74). For the two accuracy-comparable pairs, the KAN residual branch executed $13.5\times$ and $8.0\times$ slower and consumed $11.3\times$ and $5.6\times$ more energy per integration step (3.7\,$\mu$J against 0.33\,$\mu$J for the smallest pair); across all four parameter-matched size tiers the ranges are $4.7\times$--$14.5\times$ and $4.7\times$--$18.7\times$. Under INT8 quantization, KAN trajectories diverged up to $43\times$ earlier than matched MLPs; the damage traces to weight quantization, not to input-side knot-interval misassignment. These results indicate that the parameter efficiency reported for KANs does not transfer to deployment cost on scalar embedded cores, and that an MLP residual branch is the more dependable default for embedded HRPINN deployment unless specific quantization co-design is used.
The brain can rapidly perform perception, prediction, and decision-making tasks. The capabilities stem from the collective coordination of neurons. This collective activity forms a manifold structure that naturally supports information representation and prediction, yet most neuromorphic work ignores this structure. Here, we construct an artificial neural manifold based on Mott memristors, enabling accurate and robust prediction while reducing the number of required samples. By constructing an artificial neuron circuit, a bell-shaped tuning curve similar to that of biological neurons is obtained. The tuning curve converges the large-scale neuronal firing into a compact, low-dimensional manifold structure. This structure satisfies the delay embedding theorem to establish a spatiotemporal information (STI) equation, enabling rapid prediction of neural activity with small sample sizes. In addition, we introduce a memory factor to modify the STI equation, which improves prediction accuracy and robustness. We not only accurately perceive incomplete images but also predict epileptic seizures. Inspired by collective neuronal activity, Wang et al. develop a Mott-memristor-based hardware system that maps complex spike signals to population-level artificial neural dynamics, enabling accurate prediction from limited data.
Rui Wang, Guolei Liu, Saisai Wang et al.· Nature Communications· 0 citations
