Existing drift detection methods focus on designing sensitive test statistics. They treat the detection threshold as a fixed hyperparameter, set once to balance false alarms and late detections, and applied uniformly across all datasets and over time. However, maintaining model performance is the key objective from the perspective of machine learning, and we observe that model performance is highly sensitive to this threshold. This observation inspires us to investigate whether a dynamic threshold could be provably better. In this paper, we prove that a threshold that adapts over time can outperform any single fixed threshold. The main idea of the proof is that a dynamic strategy, constructed by combining the best threshold from each individual data segment, is guaranteed to outperform any single threshold that apply to all segments. Based on the theorem, we propose a Dynamic Threshold Determination algorithm. It enhances existing drift detection frameworks with a novel comparison phase to inform how the threshold should be adjusted. Extensive experiments on a wide range of synthetic and real-world datasets, including both image and tabular data, validate that our approach substantially enhances the performance of state-of-the-art drift detectors.

Detecting hallucinations in Retrieval-Augmented Generation (RAG) remains a critical reliability challenge, as ungrounded responses can have severe consequences in high-stakes applications such as clinical decision support, legal research assistants, and autonomous agents that act on retrieved evidence.Prior approaches attribute hallucinations to a binary conflict between internal knowledge stored in FFNs and the retrieved context.However, this perspective is incomplete, failing to account for the impact of other components of the LLM, such as the user query, previously generated tokens, the self token, and the Final LayerNorm adjustment.To comprehensively capture the impact of these components on hallucination detection, we propose TPA which mathematically attributes each token's probability to seven distinct sources: Query, RAG Context, Past Token, Self Token, FFN, Final LayerNorm, and Initial Embedding.This attribution quantifies how each source contributes to the generation of the next token.Specifically, we aggregate these attribution scores by Part-of-Speech (POS) tags to quantify the contribution of each model component to the generation of specific linguistic categories within a response.By leveraging these patterns, such as detecting anomalies where Nouns rely heavily on LayerNorm, TPA effectively identifies hallucinated responses.Extensive experiments on five LLMs (Llama2-7B/13B, Llama3-8B, Mistral-7B, and Qwen3-8B) demonstrate that TPA achieves state-of-the-art performance across diverse architectures.

Concept drift, characterized by unpredictable changes in data distribution over time, poses significant challenges to machine learning models in streaming data scenarios. Although error rate-based concept drift detectors are widely used, they often fail to identify drift in the early stages when the data distribution changes but error rates remain constant. This paper introduces the Prediction Uncertainty Index (PU-index), derived from the prediction uncertainty of the classifier, as a superior alternative to the error rate for drift detection. Our theoretical analysis demonstrates that: (1) The PU-index can detect drift even when error rates remain stable. (2) Any change in the error rate will lead to a corresponding change in the PU-index. These properties make the PU-index a more sensitive and robust indicator for drift detection compared to existing methods. We also propose a PU-index-based Drift Detector (PUDD) that employs a novel Adaptive PU-index Bucketing algorithm for detecting drift. Empirical evaluations on both synthetic and real-world datasets demonstrate PUDD’s efficacy in detecting drift in structured and image data.

The sample selection approach is very popular in learning with noisy labels. As deep networks "learn pattern first", prior methods built on sample selection share a similar training procedure: the small-loss examples can be regarded as clean examples and used for helping generalization, while the large-loss examples are treated as mislabeled ones and excluded from network parameter updates. However, such a procedure is arguably debatable from two folds: (a) it does not consider the bad influence of noisy labels in selected small-loss examples; (b) it does not make good use of the discarded large-loss examples, which may be clean or have meaningful information for generalization. In this paper, we propose regularly truncated M-estimators (RTME) to address the above two issues simultaneously. Specifically, RTME can alternately switch modes between truncated M-estimators and original M-estimators. The former can adaptively select small-losses examples without knowing the noise rate and reduce the side-effects of noisy labels in them. The latter makes the possibly clean examples but with large losses involved to help generalization. Theoretically, we demonstrate that our strategies are label-noise-tolerant. Empirically, comprehensive experimental results show that our method can outperform multiple baselines and is robust to broad noise types and levels.

Concept drift is an inevitable problem in non-stationary data stream environments, due to changes in data distribution over time. In practical applications, multi-stream data is more common and complex than single-stream data, yet they have received little attention. Addressing the concept drift problem while mining correlations between data streams has become a significant challenge. Several research works focus on capturing the correlation between streams using graph neural networks (GNNs), which provides valuable insights. However, these methods fix the correlation graph structure after training and are unable to adapt to the new data distribution with dynamic correlations during testing. To bridge this gap, we propose a novel concept drift self-adaptation framework based on dynamic graph regularization for multi-stream, named Multi-stream Self-adaptation based on Graph Regularization (MSGR). A new graph neural network architecture is proposed to capture deep spatio-temporal correlations and learn a correlation graph structure without any pre-defined graphs. Each node on the graph represents a stream. The correlation graph structure is constructed through Gumbel sampling and an adaptive matrix from the perspective of stream pairs. Thus we attain a high-performance GNN as the base prediction model for the multi-stream multi-step prediction task in the testing stage. To adapt to the new data distribution, we design a self-adaptation mechanism performed by assigning dynamic learning weight for newly arriving samples. Intuitively, we should assign larger learning weights for relevant samples when drift occurs. The self-adaptation process is accomplished by the sub-graph updating and the proposed graph regularization. Error-based drift detection is integrated into the framework. When drift is detected, the weight for sub-graph updating is increased by adjusting the regularization coefficient. In this way, regardless of the type and degree of concept drift occurring on one or more streams, MSGR can achieve high self-adaptation performance and provide accurate prediction results consistently. The comprehensive testing results on both real-world and synthetic datasets show that MSGR can achieve state-of-the-art performance.

Public Bike Systems (PBSs) offer convenient and green travel service and become popular around the world. In many cities, the local governments build thousands of fixed stations for PBS to alleviate the city traffic jam and solve the last-mile problem. However, the increasing use of PBSs leads to new congestion problems in the form that users have, such as no bike to rent or no dock to return the bike. Further, users wish to receive assistance on deciding how to select bike trips with minimal time cost while taking congestion into account. Meanwhile, crowdsourcing attracted increasing attention in recent years. This paper applies it to help users share information and select bike trips before the bikes or docks are occupied. An interesting and important problem is how to help users select bike trips so that the time consumed on the trips can be minimized. We model the problem as a Bike Trip Selection (BTS) game which is shown to be equivalent to the symmetric network congestion game. This equivalence allows us to design a BTS algorithm by which the users can find at least one Nash Equilibria (NE) distributively. Furthermore, this paper evaluates the algorithm based on real datasets collected from the PBS of Hangzhou City in China. We also design a BTS system including an Android APP and a server to conduct the experiment for the distributed BTS algorithm in practice.

In Bike-Sharing System (BSS), great efforts have been devoted to performing resources prediction, redistribution and trip planning to alleviate the unbalance of resources and inconvenience of bike utilization caused by the explosion of users. However, there is few work in trip planning noticing that the complete trip composes of three segments: from user's start point to a start station, from the start station to a target station and from the target station to user's terminal point. To study the case, this paper addresses a static trip planning problem in BSS by considering system-wide conflicts so as to achieve higher service quality of the system. The problem is formulated as the well-known weighted k-set packing problem. We design two algorithms, a Greedy Trip Planning algorithm (GTP) and a Humble Trip Planning algorithm (HTP), for the problem. For comparison, we design a Random Trip Planning algorithm (RTP) as a benchmark. Extensive simulation results show that GTP and HTP outperform RTP and reveal the impact of different factors on our algorithms.