Large language models (LLMs) often exhibit undesirable behaviors, such as safety violations and hallucinations.Although inference-time steering offers a cost-effective way to adjust model behavior without updating its parameters, existing methods often fail to be simultaneously effective, utilitypreserving, and training-efficient due to their rigid, one-size-fits-all designs and limited adaptability.In this work, we present FineSteer, a novel steering framework that decomposes inference-time steering into two complementary stages-conditional steering and fine-grained vector synthesis-allowing finegrained control over when and how to steer internal representations.In the first stage, we introduce a Subspace-guided Conditional Steering (SCS) mechanism that preserves model utility by avoiding unnecessary steering.In the second stage, we propose a Mixture-of-Steering-Experts (MoSE) mechanism that captures the multimodal nature of desired steering behaviors and generates query-specific steering vectors for improved effectiveness.Through tailored designs in both SCS and MoSE, FineSteer maintains robust performance on general queries while adaptively optimizing steering vectors for targeted inputs in a training-efficient manner.Extensive experiments on safety and truthfulness benchmarks show that FineSteer outperforms the state-of-the-art methods in overall performance (e.g., A 7.6% improvement on TruthfulQA over Llama-3), achieving stronger steering performance with minimal utility loss.The code is available at https://github.com/YukinoAsuna/FineSteer.

Ensuring AI safety is crucial as large language models become increasingly integrated into real-world applications.A key challenge is jailbreak, where adversarial prompts bypass built-in safeguards to elicit harmful disallowed outputs.Inspired by psychological foot-in-thedoor principles, we introduce FITD, a novel multi-turn jailbreak method that leverages the phenomenon where minor initial commitments lower resistance to more significant or more unethical transgressions.Our approach progressively escalates the malicious intent of user queries through intermediate bridge prompts and aligns the model's response by itself to induce toxic responses.Extensive experimental results on two jailbreak benchmarks demonstrate that FITD achieves an average attack success rate of 94% across seven widely used models, outperforming existing state-of-theart methods.Additionally, we provide an indepth analysis of LLM self-corruption, highlighting vulnerabilities in current alignment strategies and emphasizing the risks inherent in multi-turn interactions.The code is available at https://github.com/Jinxiaolong1129/Foot-in- the-door-Jailbreak.

In recent years, advancements in deep learning have spurred the development of numerous models for Long-term Time Series Forecasting (LTSF). However, most existing approaches struggle to fully capture the complex and structured dependencies inherent in time series data. In this work, we identify and formally define three critical dependencies that are fundamental to forecasting accuracy: order dependency and semantic dependency along the temporal dimension, as well as cross-variate dependency across the feature dimension. These dependencies are often treated in isolation, and improper handling can introduce noise and degrade forecasting performance. To bridge this gap, we investigate the potential of State Space Models (SSMs) for LTSF and emphasize their inherent advantages in capturing these essential dependencies. Additionally, we empirically observe that excessive nonlinearity in conventional SSMs introduce redundancy when applied to semantically sparse time series data. Motivated by this insight, we propose SDE (Simplified and Disentangled Dependency Encoding), a novel framework designed to enhance the capability of SSMs for LTSF. Specifically, we first eliminate unnecessary nonlinearities in vanilla SSMs, thereby improving the suitability for time series forecasting. Building on this foundation, we introduce a disentangled encoding strategy, which empowers SSMs to efficiently model cross-variate dependencies while mitigating interference between the temporal and feature dimensions. Furthermore, we provide rigorous theoretical justifications to substantiate our design choices. Extensive experiments on nine real-world benchmark datasets demonstrate that SDE-enhanced SSMs consistently outperform state-of-the-art time series forecasting models. Our code is available at https://github.com/YukinoAsuna/SAMBA.

Smart homes, powered by the Internet of Things, offer great convenience but also pose security concerns due to abnormal behaviors, such as improper operations of users and potential attacks from malicious attackers. Several behavior modeling methods have been proposed to identify abnormal behaviors and mitigate potential risks. However, their performance often falls short because they do not effectively learn less frequent behaviors, consider temporal context, or account for the impact of noise in human behaviors. In this paper, we propose SmartGuard, an autoencoder-based unsupervised user behavior anomaly detection framework. First, we design a Loss-guided Dynamic Mask Strategy (LDMS) to encourage the model to learn less frequent behaviors, which are often overlooked during learning. Second, we propose a Three-level Time-aware Position Embedding (TTPE) to incorporate temporal information into positional embedding to detect temporal context anomaly. Third, we propose a Noise-aware Weighted Reconstruction Loss (NWRL) that assigns different weights for routine behaviors and noise behaviors to mitigate the interference of noise behaviors during inference. Comprehensive experiments on three datasets with ten types of anomaly behaviors demonstrates that SmartGuard consistently outperforms state-of-the-art baselines and also offers highly interpretable results.

Node embedding learns low-dimensional vectors for nodes in the graph. Recent state-of-the-art embedding approaches take Personalized PageRank (PPR) as the proximity measure and factorize the PPR matrix or its adaptation to generate embeddings. However, little previous work analyzes what information is encoded by these approaches, and how the information correlates with their superb performance in downstream tasks. In this work, we first show that state-of-the-art embedding approaches that factorize a PPR-related matrix can be unified into a closed-form framework. Then, we study whether the embeddings generated by this strategy can be inverted to better recover the graph topology information than random-walk based embeddings. To achieve this, we propose two methods for recovering graph topology via PPR-based embeddings, including the analytical method and the optimization method. Extensive experimental results demonstrate that the embeddings generated by factorizing a PPR-related matrix maintain more topological information, such as common edges and community structures, than that generated by random walks, paving a new way to systematically comprehend why PPR-based node embedding approaches outperform random walk-based alternatives in various downstream tasks. To the best of our knowledge, this is the first work that focuses on the interpretability of PPR-based node embedding approaches.