Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks

Transformers often display an attention sink: probability mass concentrates on a fixed, content-agnostic position. We prove that computing a simple trigger-conditional behavior necessarily induces a sink in softmax self-attention models. Our results formalize a familiar intuition: normalization over a probability simplex must force attention to collapse onto a stable anchor to realize a default state (e.g., when the model needs to ignore the input). We instantiate this with a concrete task: when a designated trigger token appears, the model must return the average of all preceding token representations, and otherwise output zero, a task which mirrors the functionality of attention heads in the wild (Barbero et al., 2025; Guo et al., 2024). We also prove that non-normalized ReLU attention can solve the same task without any sink, confirming that the normalization constraint is the fundamental driver of sink behavior. Experiments validate our predictions and demonstrate they extend beyond the theoretically analyzed setting: softmax models develop strong sinks while ReLU attention eliminates them in both single-head and multi-head variants.

STAIRS-Former: Spatio-Temporal Attention with Interleaved Recursive Structure Transformer for Offline Multi-task Multi-agent Reinforcement Learning

Offline multi-agent reinforcement learning (MARL) with multi-task datasets is challenging due to varying numbers of agents across tasks and the need to generalize to unseen scenarios. Prior works employ transformers with observation tokenization and hierarchical skill learning to address these issues. However, they underutilize the transformer attention mechanism for inter-agent coordination and rely on a single history token, which limits their ability to capture long-horizon temporal dependencies in partially observable MARL settings. In this paper, we propose STAIRS-Former, a transformer architecture augmented with spatial and temporal hierarchies that enables effective attention over critical tokens while capturing long interaction histories. We further introduce token dropout to enhance robustness and generalization across varying agent populations. Extensive experiments on diverse multi-agent benchmarks, including SMAC, SMAC-v2, MPE, and MaMuJoCo, with multi-task datasets demonstrate that STAIRS-Former consistently outperforms prior methods and achieves new state-of-the-art performance.

Towards Intelligent Spectrum Management: Spectrum Demand Estimation Using Graph Neural Networks

The growing demand for wireless connectivity, combined with limited spectrum resources, calls for more efficient spectrum management. Spectrum sharing is a promising approach; however, regulators need accurate methods to characterize demand dynamics and guide allocation decisions. This paper builds and validates a spectrum demand proxy from public deployment records and uses a graph attention network in a hierarchical, multi-resolution setup (HR-GAT) to estimate spectrum demand at fine spatial scales. The model captures both neighborhood effects and cross-scale patterns, reducing spatial autocorrelation and improving generalization. Evaluated across five Canadian cities and against eight competitive baselines, HR-GAT reduces median RMSE by roughly 21% relative to the best alternative and lowers residual spatial bias. The resulting demand maps are regulator-accessible and support spectrum sharing and spectrum allocation in wireless networks.

Telogenesis: Goal Is All U Need

Goal-conditioned systems assume goals are provided externally. We ask whether attentional priorities can emerge endogenously from an agent's internal cognitive state. We propose a priority function that generates observation targets from three epistemic gaps: ignorance (posterior variance), surprise (prediction error), and staleness (temporal decay of confidence in unobserved variables). We validate this in two systems: a minimal attention-allocation environment (2,000 runs) and a modular, partially observable world (500 runs). Ablation shows each component is necessary. A key finding is metric-dependent reversal: under global prediction error, coverage-based rotation wins; under change detection latency, priority-guided allocation wins, with advantage growing monotonically with dimensionality (d = -0.95 at N=48, p < 10^-6). Detection latency follows a power law in attention budget, with a steeper exponent for priority-guided allocation (0.55 vs. 0.40). When the decay rate is made learnable per variable, the system spontaneously recovers environmental volatility structure without supervision (t = 22.5, p < 10^-6). We demonstrate that epistemic gaps alone, without external reward, suffice to generate adaptive priorities that outperform fixed strategies and recover latent environmental structure.

Attention Gathers, MLPs Compose: A Causal Analysis of an Action-Outcome Circuit in VideoViT

The paper explores how video models trained for classification tasks represent nuanced, hidden semantic information that may not affect the final outcome, a key challenge for Trustworthy AI models. Through Explainable and Interpretable AI methods, specifically mechanistic interpretability techniques, the internal circuit responsible for representing the action's outcome is reverse-engineered in a pre-trained video vision transformer, revealing that the "Success vs Failure" signal is computed through a distinct amplification cascade. While there are low-level differences observed from layer 0, the abstract and semantic representation of the outcome is progressively amplified from layers 5 through 11. Causal analysis, primarily using activation patching supported by ablation results, reveals a clear division of labor: Attention Heads act as "evidence gatherers", providing necessary low-level information for partial signal recovery, while MLP Blocks function as robust "concept composers", each of which is the primary driver to generate the "success" signal. This distributed and redundant circuit in the model's internals explains its resilience to simple ablations, demonstrating a core computational pattern for processing human-action outcomes. Crucially, the existence of this sophisticated circuit for representing complex outcomes, even within a model trained only for simple classification, highlights the potential for models to develop forms of 'hidden knowledge' beyond their explicit task, underscoring the need for mechanistic oversight for building genuinely Explainable and Trustworthy AI systems intended for deployment.

Alfa: Attentive Low-Rank Filter Adaptation for Structure-Aware Cross-Domain Personalized Gaze Estimation

Pre-trained gaze models learn to identify useful patterns commonly found across users, but subtle user-specific variations (i.e., eyelid shape or facial structure) can degrade model performance. Test-time personalization (TTP) adapts pre-trained models to these user-specific domain shifts using only a few unlabeled samples. Efficient fine-tuning is critical in performing this domain adaptation: data and computation resources can be limited-especially for on-device customization. While popular parameter-efficient fine-tuning (PEFT) methods address adaptation costs by updating only a small set of weights, they may not be taking full advantage of structures encoded in pre-trained filters. To more effectively leverage existing structures learned during pre-training, we reframe personalization as a process to reweight existing features rather than learning entirely new ones. We present Attentive Low-Rank Filter Adaptation (Alfa) to adapt gaze models by reweighting semantic patterns in pre-trained filters. With Alfa, singular value decomposition (SVD) extracts dominant spatial components that capture eye and facial characteristics across users. Via an attention mechanism, we need only a few unlabeled samples to adjust and reweight pre-trained structures, selectively amplifying those relevant to a target user. Alfa achieves the lowest average gaze errors across four cross-dataset gaze benchmarks, outperforming existing TTP methods and low-rank adaptation (LoRA)-based variants. We also show that Alfa's attentive low-rank methods can be applied to applications beyond vision, such as diffusion-based language models.

SCDP: Learning Humanoid Locomotion from Partial Observations via Mixed-Observation Distillation

Distilling humanoid locomotion control from offline datasets into deployable policies remains a challenge, as existing methods rely on privileged full-body states that require complex and often unreliable state estimation. We present Sensor-Conditioned Diffusion Policies (SCDP) that enables humanoid locomotion using only onboard sensors, eliminating the need for explicit state estimation. SCDP decouples sensing from supervision through mixed-observation training: diffusion model conditions on sensor histories while being supervised to predict privileged future state-action trajectories, enforcing the model to infer the motion dynamics under partial observability. We further develop restricted denoising, context distribution alignment, and context-aware attention masking to encourage implicit state estimation within the model and to prevent train-deploy mismatch. We validate SCDP on velocity-commanded locomotion and motion reference tracking tasks. In simulation, SCDP achieves near-perfect success on velocity control (99-100%) and 93% tracking success in AMASS test set, performing comparable to privileged baselines while using only onboard sensors. Finally, we deploy the trained policy on a real G1 humanoid at 50 Hz, demonstrating robust real robot locomotion without external sensing or state estimation.

Beyond Hungarian: Match-Free Supervision for End-to-End Object Detection

Recent DEtection TRansformer (DETR) based frameworks have achieved remarkable success in end-to-end object detection. However, the reliance on the Hungarian algorithm for bipartite matching between queries and ground truths introduces computational overhead and complicates the training dynamics. In this paper, we propose a novel matching-free training scheme for DETR-based detectors that eliminates the need for explicit heuristic matching. At the core of our approach is a dedicated Cross-Attention-based Query Selection (CAQS) module. Instead of discrete assignment, we utilize encoded ground-truth information to probe the decoder queries through a cross-attention mechanism. By minimizing the weighted error between the queried results and the ground truths, the model autonomously learns the implicit correspondences between object queries and specific targets. This learned relationship further provides supervision signals for the learning of queries. Experimental results demonstrate that our proposed method bypasses the traditional matching process, significantly enhancing training efficiency, reducing the matching latency by over 50\%, effectively eliminating the discrete matching bottleneck through differentiable correspondence learning, and also achieving superior performance compared to existing state-of-the-art methods.

BuildMamba: A Visual State-Space Based Model for Multi-Task Building Segmentation and Height Estimation from Satellite Images

Accurate building segmentation and height estimation from single-view RGB satellite imagery are fundamental for urban analytics, yet remain ill-posed due to structural variability and the high computational cost of global context modeling. While current approaches typically adapt monocular depth architectures, they often suffer from boundary bleeding and systematic underestimation of high-rise structures. To address these limitations, we propose BuildMamba, a unified multi-task framework designed to exploit the linear-time global modeling of visual state-space models. Motivated by the need for stronger structural coupling and computational efficiency, we introduce three modules: a Mamba Attention Module for dynamic spatial recalibration, a Spatial-Aware Mamba-FPN for multi-scale feature aggregation via gated state-space scans, and a Mask-Aware Height Refinement module using semantic priors to suppress height artifacts. Extensive experiments demonstrate that BuildMamba establishes a new performance upper bound across three benchmarks. Specifically, it achieves an IoU of 0.93 and RMSE of 1.77~m on DFC23 benchmark, surpassing state-of-the-art by 0.82~m in height estimation. Simulation results confirm the model's superior robustness and scalability for large-scale 3D urban reconstruction.

Benchmark of Benchmarks: Unpacking Influence and Code Repository Quality in LLM Safety Benchmarks

The rapid growth of research in LLM safety makes it hard to track all advances. Benchmarks are therefore crucial for capturing key trends and enabling systematic comparisons. Yet, it remains unclear why certain benchmarks gain prominence, and no systematic assessment has been conducted on their academic influence or code quality. This paper fills this gap by presenting the first multi-dimensional evaluation of the influence (based on five metrics) and code quality (based on both automated and human assessment) on LLM safety benchmarks, analyzing 31 benchmarks and 382 non-benchmarks across prompt injection, jailbreak, and hallucination. We find that benchmark papers show no significant advantage in academic influence (e.g., citation count and density) over non-benchmark papers. We uncover a key misalignment: while author prominence correlates with paper influence, neither author prominence nor paper influence shows a significant correlation with code quality. Our results also indicate substantial room for improvement in code and supplementary materials: only 39% of repositories are ready-to-use, 16% include flawless installation guides, and a mere 6% address ethical considerations. Given that the work of prominent researchers tends to attract greater attention, they need to lead the effort in setting higher standards.