Generative recommenders represent items as short discrete Semantic IDs (SIDs) that are decoded autoregressively, yet most prior work derives these codes by quantising isolated content embeddings. This paper argues that SID construction is fundamentally a recursive clustering problem over a collaborative item graph, and proposes GrIS, a framework that separates graph construction from hierarchical partition to balance semantic and behavioural signals. By exposing these two axes, the authors show that content-only tokenisers are merely empty-graph instances of a broader design space, and they demonstrate new instantiations that improve SID quality for downstream recommendation.

The paper tackles a critical problem in automated image-editing evaluation: multimodal large language models (MLLMs) acting as judges may be swayed by irrelevant presentation cues rather than the edit itself. Because visual interventions can alter the very quality being assessed, the authors introduce EditJudgeBias, a counterfactual benchmark of 1,196 real samples where 13 cues are injected across four evaluation stages and rigorously verified to preserve underlying quality. They audit five judges along three dimensions—invariance to cues, agreement with humans, and pairwise stability—and show that every judge shifts beyond its own noise floor, with fabricated social proof inflating scores and display-order swaps reversing up to 60.9% of verdicts.

Predictive maintenance models are typically tuned to maximize $R^2$, a symmetric metric that ignores the fact that late remaining useful life (RUL) predictions are far more costly than early ones. This study treats the optimization objective itself as a variable, comparing single-objective and multi-objective hyperparameter optimization across MLP, LSTM, XGBoost, TCN, and Transformer on NASA C-MAPSS and BackBlaze benchmarks. The core result is that NSGA-II with Entropy–CRITIC selection cuts directional imbalance by roughly a third on simulated data without sacrificing accuracy, though this advantage collapses into near-random behavior on real-world hard-drive data.

This paper tackles privacy, energy, and infrastructure constraints in smart buildings by proposing a decentralized cloud-fog-edge architecture that treats the building itself as a data center. Its core idea couples the KOptim container orchestrator with FIWARE for dynamic SLA management, while running lightweight online ML algorithms such as GHA-PCA and K-means directly on Arduino-class microcontrollers. The humanitarian motivation—bringing autonomous AI to underserved communities—is compelling, although the current evaluation remains largely preliminary.

This workshop paper examines how fairness assessments for ICU mortality models on MIMIC-IV hinge on both the chosen performance metric and the demographic resolution of evaluation. It introduces a lightweight classifier-head adaptation, IBHFT, that jointly balances ethnicity–gender–insurance representation without conditioning on mortality outcomes, using this case study to show that marginal summaries and single-metric reports can paint a misleading picture of clinical fairness. The core argument matters because deploying models based on aggregate AUROC or one-axis parity alone can obscure dangerous error-rate disparities at clinically consequential intersections.

pCoMole addresses the problem of editing existing biomolecular sequences to improve multiple properties while obeying hard feasibility constraints such as PAM specificity or emission range. Its core idea is to steer a pre-trained Edit Flow via a Doob-$h$ transform, using an augmented Tchebycheff utility over terminal states that is approximated through short Monte Carlo rollouts. This matters because existing shrinkage methods like RayGun and SCISOR optimize length alone, whereas biomolecular engineering inherently requires trading off function, developability, and manufacturability. The work is notable for combining a clean theoretical justification with preliminary wet-lab validation on shortened eGFP variants.

This paper asks whether a fixed offline corpus generated by a behaviour policy can ever distinguish two candidate Regular Decision Processes (RDPs). It proves that when two candidates are $\pi$-observationally equivalent, their posterior odds remain frozen at the prior for any sample size (Theorem 6), gives an exact measure-zero characterization via the product-automaton blame language (Theorem 12), and derives PEC, a linear-time algorithm that certifies equivalence and names the missing experiment. The work matters because existing PAC offline guarantees for RDPs assume a distinguishability condition that was previously impossible to verify.

This paper develops a high-dimensional theory for why masked autoencoders learn better representations than standard autoencoders. In a spiked model with heterogeneous noise, the authors prove that a linear masked autoencoder recovers the latent feature at linear sample complexity $n=O(d)$ where PCA fails. They further quantify how dynamic mask resampling lowers sample complexity compared to static masking, and validate the qualitative predictions in CNNs, vision transformers, and a BERT pilot.

This paper proves worst-case lower bounds for stochastic first-order algorithms in nonconvex--concave minimax optimization, explicitly permitting variance-reduction via same-sample oracle queries. Its main result shows that the stochastic barrier remains $\Omega(L^3 D_Y^2 \Delta \sigma^2 \varepsilon^{-6})$ even with mean-square smoothness. Complementary nonconvex--strongly-concave bounds are also provided. Together the results establish that variance reduction does not improve the fundamental accuracy exponent in the minimax setting.

This paper studies stochastic multi-level compositional optimization, where the objective is a nested composition of $T$ smooth non-convex functions and only stochastic per-layer function and gradient values are available. The core idea is to use momentum-based estimators to track inner function values level-by-level, and then apply the chain rule to build a momentum gradient estimator for the overall objective. It matters because it establishes that the optimal $\mathcal{O}(\epsilon^{-4})$ sample complexity is achievable under standard smoothness assumptions, closing a gap with single-level lower bounds without relying on the stronger average-smoothness conditions common in prior multi-level literature.

Unmanned aerial vehicle swarms operating with independent edge-deployed language models face a dual challenge: reasoning context degrades as interaction histories grow, and indiscriminate information sharing wastes limited bandwidth. This paper proposes a distributed architecture in which each UAV maintains structured atomic notes in bounded core, local, and peer-specific memory queues, coupled with a deterministic Interest-Aware Gossip Engine that forwards messages based on semantic novelty and recency. The work matters because it targets the practical gap between centralized LLM coordination, which introduces bottlenecks and single points of failure, and fully decentralized control, which risks context degradation and communication collapse.

Federated learning on edge devices must simultaneously handle varying resource budgets and domain-shifted local data. FedSAP frames this coupled challenge as a structured pruning problem where channels are assigned to Global, Private, or Dropped states. By inferring pseudo-domains from shallow gradient similarity and allocating layer-wise budgets adaptively, the method improves global mean accuracy under aggressive heterogeneous compression.

Explaining why neural networks generalize and how training unfolds remains a central challenge in deep learning. This paper introduces Magnitude Potential (MP), a point-set coverage measure from metric space theory, and defines the MP ratio $\mathcal{R}^t_{X,X_c}(z)$ as the within-class potential divided by the full-dataset potential at the logit layer. The authors argue this single geometric score predicts per-sample memorization, detects decision-boundary phase transitions, and foreshadows grokking without requiring model retraining or architectural assumptions.

LITL addresses the problem of recovering target Langevin dynamics when only biased or static source samples are available, complemented by black-box evaluative feedback. The core idea is to learn the dominant spectral subspace of the target infinitesimal generator and its projected drift via importance-weighted Dirichlet representation learning, explicitly avoiding target trajectories, gradients, or retraining. This matters for molecular dynamics, auditing generative models, and post-hoc steering of learned representations under distribution shift.

Deep learning audio research hides a massive energy footprint behind the final training run. This paper develops a methodology to estimate the full development-phase cost of audio DL projects by mining job logs from the Grid5000 shared computing platform. By tracing thousands of individual jobs across four representative projects, the authors show that prototyping, debugging, and hyper-parameter search can consume $3 imes$ to $256 imes$ more energy than training the reported best models, calling for systematic lifecycle reporting.

Hob-VL introduces a diagnostic benchmark for visually grounded Boolean reasoning, asking vision-language models to evaluate ten-literal Boolean combinations over images and to identify objects satisfying Boolean descriptions. The core idea is to pair each expression with a mechanically verified De Morgan equivalent and to test both symbolic and natural-language presentations, isolating compositional reasoning from surface memorization. It matters because real-world visual queries—warehouse checks, agricultural monitoring, medical imaging—depend on reliably combining multiple visual facts under logical constraints.

Decision models that score a variable-sized set of candidate actions from a serialized sequence are vulnerable to permutation sensitivity because standard causal attention lets later candidates attend to earlier ones and uses absolute positions. The paper proposes candidate-independent block-causal attention (CI), which blocks cross-candidate information flow and resets per-candidate positions so that each candidate is scored based only on the shared prefix and its own tokens. This matters for System 1 components inside generative systems, where runtime-dependent candidate order should not become a latent feature; the authors show that CI cuts flip rates by an order of magnitude across three backbones while preserving competitive accuracy.

Iterative policy improvement is central to practical RL, but approximate value estimates make greedification unreliable. This paper defines optimal policy improvement restricted to a state subset as solving an induced MDP, and derives $\mathcal{I}_{\text{opt}}$, a novel greedification operator that formulates approximate evaluation as probabilistic decision-making under uncertainty. It weights advantages by action-dependent uncertainty, generalizes popular regularized operators, and improves aggregate performance across model-based and model-free, discrete and continuous, and online and offline settings.

This paper extends one-step generative modeling to finite discrete spaces by replacing Euclidean drift with a discrete Wasserstein gradient flow. Using Maas geometry on a reversible Markov kernel, it defines a relative-KL flow that is realized at the particle level by Markov jumps and amortized into a neural generator. The result is a theoretically grounded analogue of continuous drifting that preserves one-step inference. The work is validated only on a small controlled torus, leaving real-world scalability as open future work.

AURAL tackles the tension between reasoning quality and low latency in speech language models by replacing explicit chain-of-thought with adaptive latent reasoning. Instead of generating slow text intermediates, the model predicts chunks of continuous latent states drawn from a Gaussian mixture, cutting time to the first token by more than 11x while largely matching explicit CoT accuracy across two backbones. The work also constructs AuralReason-683K, a bilingual corpus of 683K utterances with concise reasoning traces used to initialize latent training.