Probing Routing-Conditional Calibration in Attention-Residual Transformers

Post-hoc calibration is usually evaluated as a function of logits or softmax confidence alone, even as routing-augmented architectures increasingly accompany predictions with sample-specific internal routing traces and pair them with claims of calibration-relevant uncertainty. We ask a basic question: do these traces provide stable routing-specific evidence for post-hoc calibration beyond confidence? We study this in Attention-Residual transformers (Kimi Team, 2026) through a matched-confidence diagnostic suite that stratifies examples by routing-derived state, compares subgroup gaps against within-bin routing-permutation nulls, and evaluates matched post-hoc probes differing only in their auxiliary feature. Across our completed AR runs, scalar routing summaries do not provide stable evidence of routing-conditional miscalibration: weighted gaps remain small or seed-sensitive, and only $1$ of $30$ within-bin permutation tests rejects the conditional-null at $α=0.05$ (only on one seed; not stable across seeds in that cell). AR-CondCal, a minimal $2$-D Nadaraya--Watson probe on confidence and routing-depth variance, lies within the seed-variance band of matched confidence-only and predictive-entropy controls and does not reliably improve worst-routing-tertile ECE; bandwidth-sensitivity checks (Scott multiples, CV-NLL, global-ECE oracle) do not change this. A full-vector MLP over $(c, H_1, \ldots, H_L)$ can appear to improve over a linear confidence baseline, but the apparent gain disappears once a capacity-matched confidence-only MLP is included as a control, and shuffled routing profiles achieve comparable performance. Apparent routing-aware calibration gains in this AR setting should not be read as internal-state calibration until matched-confidence, bandwidth, capacity, and permutation controls rule out common confounds.

Beyond Fine-Tuning: In-Context Learning and Chain-of-Thought for Reasoned Distractor Generation

Distractor generation (DG) remains a labor-intensive task that still significantly depends on domain experts. The task focuses on generating plausible yet incorrect options, known as distractors, for multiple-choice questions. A reliable distractor must be contextually relevant to the question and able to mislead examinees through implicit reasoning when identifying the correct answer. While a recent method integrates fine-tuning pre-trained encoder-decoder models with contrastive learning to generate semantically relevant distractors for a given question-answer, it often fails to capture the underlying reasoning process that experts utilize when selecting distractors in benchmarks. In this paper, we explore large language models (LLMs) reasoning for DG through in-context learning with unsupervised semantic retrieval for selecting few-shot examples. We design a rationale-augmented DG framework that jointly generates distractors and their rationales for a given question-answer. Extensive experiments on six benchmarks, with varying average distractor lengths and domains, demonstrate that prompting LLMs with few-shot examples substantially improves the performance compared to recent DG models. It outperforms recent approaches and achieves state-of-the-art results in generating reasoned distractors that align with human-labeled benchmarks.

Test-Time Attention Purification for Backdoored Large Vision Language Models

Despite the strong multimodal performance, large vision-language models (LVLMs) are vulnerable during fine-tuning to backdoor attacks, where adversaries insert trigger-embedded samples into the training data to implant behaviors that can be maliciously activated at test time. Existing defenses typically rely on retraining backdoored parameters (e.g., adapters or LoRA modules) with clean data, which is computationally expensive and often degrades model performance. In this work, we provide a new mechanistic understanding of backdoor behaviors in LVLMs: the trigger does not influence prediction through low-level visual patterns, but through abnormal cross-modal attention redistribution, where trigger-bearing visual tokens steal attention away from the textual context - a phenomenon we term attention stealing. Motivated by this, we propose CleanSight, a training-free, plug-and-play defense that operates purely at test time. CleanSight (i) detects poisoned inputs based on the relative visual-text attention ratio in selected cross-modal fusion layers, and (ii) purifies the input by selectively pruning the suspicious high-attention visual tokens to neutralize the backdoor activation. Extensive experiments show that CleanSight significantly outperforms existing pixel-based purification defenses across diverse datasets and backdoor attack types, while preserving the model's utility on both clean and poisoned samples.

Calibration Attention: Instance-wise Temperature Scaling for Vision Transformers

Probability calibration is critical when Vision Transformers are deployed in risk-sensitive applications. The standard fix, post-hoc temperature scaling, uses a single global scalar and requires a held-out validation set. We introduce Calibration Attention (CalAttn), a drop-in module that learns an adaptive, per-instance temperature directly from the ViT's CLS token. Across CIFAR-10/100, MNIST, Tiny-ImageNet, and ImageNet-1K, CalAttn reduces calibration error by up to 4x on ViT-224, DeiT, and Swin, while adding under 0.1 percent additional parameters. The learned temperatures cluster tightly around 1.0, in contrast to the large global values used by standard temperature scaling. CalAttn is simple, efficient, and architecture-agnostic, and yields more trustworthy probabilities without sacrificing accuracy. Code: [https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-](https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-)

Rethinking Gating Mechanism in Sparse MoE: Handling Arbitrary Modality Inputs with Confidence-Guided Gate

Effectively managing missing modalities is a fundamental challenge in real-world multimodal learning scenarios, where data incompleteness often results from systematic collection errors or sensor failures. Sparse Mixture-of-Experts (SMoE) architectures have the potential to naturally handle multimodal data, with individual experts specializing in different modalities. However, existing SMoE approach often lacks proper ability to handle missing modality, leading to performance degradation and poor generalization in real-world applications. We propose Conf-SMoE to introduce a two-stage imputation module to handle the missing modality problem for the SMoE architecture and reveal the insight of expert collapse from theoretical analysis with strong empirical evidence. Inspired by our theoretical analysis, Conf-SMoE propose a novel expert gating mechanism by detaching the softmax routing score to task confidence score w.r.t ground truth. This naturally relieves expert collapse without introducing additional load balance loss function. We show that the insights of expert collapse aligns with other gating mechanism such as Gaussian and Laplacian gate. We also evaluate the proposed method on four different real world dataset with three different experiment settings to conduct comprehensive the analysis of Conf-SMoE on modality fusion and resistance to missing modality.

Electricity Cost Minimization for Multi-Workflow Allocation in Geo-Distributed Data Centers

Worldwide, Geo-distributed Data Centers (GDCs) provide computing and storage services for massive workflow applications, resulting in high electricity costs that vary depending on geographical locations and time. How to reduce electricity costs while satisfying the deadline constraints of workflow applications is important in GDCs, which is determined by the execution time of servers, power, and electricity price. Determining the completion time of workflows with different server frequencies can be challenging, especially in scenarios with heterogeneous computing resources in GDCs. Moreover, the electricity price is also different in geographical locations and may change dynamically. To address these challenges, we develop a geo-distributed system architecture and propose an Electricity Cost aware Multiple Workflows Scheduling algorithm (ECMWS) for servers of GDCs with fixed frequency and power. ECMWS comprises four stages, namely workflow sequencing, deadline partitioning, task sequencing, and resource allocation where two graph embedding models and a policy network are constructed to solve the Markov Decision Process (MDP). After statistically calibrating parameters and algorithm components over a comprehensive set of workflow instances, the proposed algorithms are compared with the state-of-the-art methods over two types of workflow instances. The experimental results demonstrate that our proposed algorithm significantly outperforms other algorithms, achieving an improvement of over 15\% while maintaining an acceptable computational time. The source codes are available at https://gitee.com/public-artifacts/ecmws-experiments.

PostHoc FREE Calibrating on Kolmogorov Arnold Networks

Kolmogorov Arnold Networks (KANs) are neural architectures inspired by the Kolmogorov Arnold representation theorem that leverage B Spline parameterizations for flexible, locally adaptive function approximation. Although KANs can capture complex nonlinearities beyond those modeled by standard MultiLayer Perceptrons (MLPs), they frequently exhibit miscalibrated confidence estimates manifesting as overconfidence in dense data regions and underconfidence in sparse areas. In this work, we systematically examine the impact of four critical hyperparameters including Layer Width, Grid Order, Shortcut Function, and Grid Range on the calibration of KANs. Furthermore, we introduce a novel TemperatureScaled Loss (TSL) that integrates a temperature parameter directly into the training objective, dynamically adjusting the predictive distribution during learning. Both theoretical analysis and extensive empirical evaluations on standard benchmarks demonstrate that TSL significantly reduces calibration errors, thereby improving the reliability of probabilistic predictions. Overall, our study provides actionable insights into the design of spline based neural networks and establishes TSL as a robust loss solution for enhancing calibration.

Free-Knots Kolmogorov-Arnold Network: On the Analysis of Spline Knots and Advancing Stability

Kolmogorov-Arnold Neural Networks (KANs) have gained significant attention in the machine learning community. However, their implementation often suffers from poor training stability and heavy trainable parameter. Furthermore, there is limited understanding of the behavior of the learned activation functions derived from B-splines. In this work, we analyze the behavior of KANs through the lens of spline knots and derive the lower and upper bound for the number of knots in B-spline-based KANs. To address existing limitations, we propose a novel Free Knots KAN that enhances the performance of the original KAN while reducing the number of trainable parameters to match the trainable parameter scale of standard Multi-Layer Perceptrons (MLPs). Additionally, we introduce new a training strategy to ensure $C^2$ continuity of the learnable spline, resulting in smoother activation compared to the original KAN and improve the training stability by range expansion. The proposed method is comprehensively evaluated on 8 datasets spanning various domains, including image, text, time series, multimodal, and function approximation tasks. The promising results demonstrates the feasibility of KAN-based network and the effectiveness of proposed method.

Devil in the Tail: A Multi-Modal Framework for Drug-Drug Interaction Prediction in Long Tail Distinction

Drug-drug interaction (DDI) identification is a crucial aspect of pharmacology research. There are many DDI types (hundreds), and they are not evenly distributed with equal chance to occur. Some of the rarely occurred DDI types are often high risk and could be life-critical if overlooked, exemplifying the long-tailed distribution problem. Existing models falter against this distribution challenge and overlook the multi-faceted nature of drugs in DDI prediction. In this paper, a novel multi-modal deep learning-based framework, namely TFDM, is introduced to leverage multiple properties of a drug to achieve DDI classification. The proposed framework fuses multimodal features of drugs, including graph-based, molecular structure, Target and Enzyme, for DDI identification. To tackle the challenge posed by the distribution skewness across categories, a novel loss function called Tailed Focal Loss is introduced, aimed at further enhancing the model performance and address gradient vanishing problem of focal loss in extremely long-tailed dataset. Intensive experiments over 4 challenging long-tailed dataset demonstrate that the TFMD outperforms the most recent SOTA methods in long-tailed DDI classification tasks. The source code is released to reproduce our experiment results: https://github.com/IcurasLW/TFMD_Longtailed_DDI.git

Revisited Large Language Model for Time Series Analysis through Modality Alignment

Large Language Models have demonstrated impressive performance in many pivotal web applications such as sensor data analysis. However, since LLMs are not designed for time series tasks, simpler models like linear regressions can often achieve comparable performance with far less complexity. In this study, we perform extensive experiments to assess the effectiveness of applying LLMs to key time series tasks, including forecasting, classification, imputation, and anomaly detection. We compare the performance of LLMs against simpler baseline models, such as single-layer linear models and randomly initialized LLMs. Our results reveal that LLMs offer minimal advantages for these core time series tasks and may even distort the temporal structure of the data. In contrast, simpler models consistently outperform LLMs while requiring far fewer parameters. Furthermore, we analyze existing reprogramming techniques and show, through data manifold analysis, that these methods fail to effectively align time series data with language and display pseudo-alignment behaviour in embedding space. Our findings suggest that the performance of LLM-based methods in time series tasks arises from the intrinsic characteristics and structure of time series data, rather than any meaningful alignment with the language model architecture.