Kalman Delta Networks (KDNs) extend linear attention models by treating associative memory as a linear–Gaussian state‑space system, enabling the Kalman filter to optimally estimate both memory state and its uncertainty. Two GPU‑friendly approximations—Diagonal KDN and Isotropic KDN—use mean‑field variational inference or a single scalar uncertainty per head, respectively, to maintain tractable uncertainty recurrences during linear‑attention scans. Experiments on 750 M and 1.3 B‑parameter models show that KDN variants consistently lower perplexity and raise downstream accuracy compared to existing linear‑attention baselines.
arXiv:2602. 10743v2 Announce Type: replace Abstract: State-space language models such as Mamba and gated linear attention (GLA) offer linear-complexity, parallelisable alternatives to transformers, but their linear state updates limit expressivity and robust state tracking.
arXiv:2607. 21000v1 Announce Type: new Abstract: Long-sequence memory tracking places two opposing demands on a recurrent state: near-lossless retention of stored bindings over long horizons, and active overwriting of stale ones.
The paper proposes treating a neural network’s layers as time steps in a state‑space model, converting Bayesian training into a smoothing problem. By propagating Gaussian moments forward and applying a Rauch–Tung–Striebel backward pass, weight posteriors are updated in closed form without gradient iterations or replay. The authors extend prior work by introducing a cross‑covariance identity that allows full‑covariance propagation through nonlinear activations, enabling more accurate online adaptation in non‑stationary classification, dynamics learning, and vision‑language‑action policy adaptation.
Long-sequence memory tracking places two opposing demands on a recurrent state: near-lossless retention of stored bindings over long horizons, and active overwriting of stale ones. In our diagnostic suite, the strongest efficient baselines tend to solve only one side well.
Fast Weight Attention for Continual Learning introduces recurrent fast‑weight memories and selective state‑space models that compress expanding context into a fixed‑size recurrent state, enabling an online learning rule for state transitions. The paper derives normalized first‑order updates for squared‑error regression and negative inner‑product objectives, presenting several variants (Falcon‑1, Falcon‑2, Falcon‑3 and their inner‑product counterparts) with recurrent, masked‑parallel, and chunk‑parallel implementations. These methods demonstrate competitive performance in language modeling and improved length extrapolation on variable‑digit addition tasks.
arXiv:2608. 13621v1 Announce Type: new Abstract: A hidden Markov model (HMM) combines three roles: inference of a hidden-state belief from observations, propagation through a Markov transition, and emission back to observation space.
arXiv:2605. 31163v2 Announce Type: replace-cross Abstract: We introduce the design-model framework: a way to derive efficient recurrent sequence maps from explicit assumptions about memory.
arXiv:2604.19021v3 Announce Type: replace Abstract: Linear attention mechanisms have emerged as promising alternatives to softmax attention, offering linear-time complexity during inference. Recent a...
arXiv:2503. 18970v4 Announce Type: replace Abstract: Structured State Space Models (SSMs) have become a prominent class of sequence models, developed against two long-standing difficulties: the sequential computation and gradient propagation limits of Recurrent Neural Networks (RNNs), and the quadratic time and memory cost of self-attention in Transformers.
arXiv:2606. 01468v1 Announce Type: cross Abstract: Due to their explicit priors and ability to model uncertainty, Bayesian methods have played a major role in dynamical latent variable modeling of single-cell neural recordings.
The paper introduces DAMP, a decay‑aware mixed‑precision quantization scheme for recurrent‑state representations in GDN and KDA language models. By identifying high‑risk channels through quantization‑error energy and decay persistence, DAMP stores these channels at higher precision while compressing the rest to INT8, achieving a 9.9‑bit average precision. Experiments on Qwen3.6‑35B and Kimi‑Linear‑48B show a 69.1% reduction in recurrent‑state storage, up to 2.01× faster state‑update kernels, and up to 10.9% lower full‑model TPOT while preserving accuracy close to the FP32 baseline.
arXiv:2609.36529v1 Announce Type: new Abstract: Recurrent neural networks (RNNs) compress the historical context into a memory state of fixed size, thus allowing for constant-time inference. The memo...