Backtracking for Safety

Large language models (LLMs) have demonstrated remarkable capabilities across various tasks, but ensuring their safety and alignment with human values remains crucial. Current safety alignment methods, such as supervised fine-tuning and reinforcement learning-based approaches, can exhibit vulnerabilities to adversarial attacks and often result in shallow safety alignment, primarily focusing on preventing harmful content in the initial tokens of the generated output. While methods like resetting can help recover from unsafe generations by discarding previous tokens and restarting the generation process, they are not well-suited for addressing nuanced safety violations like toxicity that may arise within otherwise benign and lengthy generations. In this paper, we propose a novel backtracking method designed to address these limitations. Our method allows the model to revert to a safer generation state, not necessarily at the beginning, when safety violations occur during generation. This approach enables targeted correction of problematic segments without discarding the entire generated text, thereby preserving efficiency. We demonstrate that our method dramatically reduces toxicity appearing through the generation process with minimal impact to efficiency.

Robust Gymnasium: A Unified Modular Benchmark for Robust Reinforcement Learning

Driven by inherent uncertainty and the sim-to-real gap, robust reinforcement learning (RL) seeks to improve resilience against the complexity and variability in agent-environment sequential interactions. Despite the existence of a large number of RL benchmarks, there is a lack of standardized benchmarks for robust RL. Current robust RL policies often focus on a specific type of uncertainty and are evaluated in distinct, one-off environments. In this work, we introduce Robust-Gymnasium, a unified modular benchmark designed for robust RL that supports a wide variety of disruptions across all key RL components-agents' observed state and reward, agents' actions, and the environment. Offering over sixty diverse task environments spanning control and robotics, safe RL, and multi-agent RL, it provides an open-source and user-friendly tool for the community to assess current methods and foster the development of robust RL algorithms. In addition, we benchmark existing standard and robust RL algorithms within this framework, uncovering significant deficiencies in each and offering new insights.

TimeDistill: Efficient Long-Term Time Series Forecasting with MLP via Cross-Architecture Distillation

Transformer-based and CNN-based methods demonstrate strong performance in long-term time series forecasting. However, their high computational and storage requirements can hinder large-scale deployment. To address this limitation, we propose integrating lightweight MLP with advanced architectures using knowledge distillation (KD). Our preliminary study reveals different models can capture complementary patterns, particularly multi-scale and multi-period patterns in the temporal and frequency domains. Based on this observation, we introduce TimeDistill, a cross-architecture KD framework that transfers these patterns from teacher models (e.g., Transformers, CNNs) to MLP. Additionally, we provide a theoretical analysis, demonstrating that our KD approach can be interpreted as a specialized form of mixup data augmentation. TimeDistill improves MLP performance by up to 18.6%, surpassing teacher models on eight datasets. It also achieves up to 7X faster inference and requires 130X fewer parameters. Furthermore, we conduct extensive evaluations to highlight the versatility and effectiveness of TimeDistill.

Reward-Safety Balance in Offline Safe RL via Diffusion Regularization

Constrained reinforcement learning (RL) seeks high-performance policies under safety constraints. We focus on an offline setting where the agent has only a fixed dataset -- common in realistic tasks to prevent unsafe exploration. To address this, we propose Diffusion-Regularized Constrained Offline Reinforcement Learning (DRCORL), which first uses a diffusion model to capture the behavioral policy from offline data and then extracts a simplified policy to enable efficient inference. We further apply gradient manipulation for safety adaptation, balancing the reward objective and constraint satisfaction. This approach leverages high-quality offline data while incorporating safety requirements. Empirical results show that DRCORL achieves reliable safety performance, fast inference, and strong reward outcomes across robot learning tasks. Compared to existing safe offline RL methods, it consistently meets cost limits and performs well with the same hyperparameters, indicating practical applicability in real-world scenarios.

Time-VLM: Exploring Multimodal Vision-Language Models for Augmented Time Series Forecasting

Recent advancements in time series forecasting have explored augmenting models with text or vision modalities to improve accuracy. While text provides contextual understanding, it often lacks fine-grained temporal details. Conversely, vision captures intricate temporal patterns but lacks semantic context, limiting the complementary potential of these modalities. To address this, we propose Time-VLM, a novel multimodal framework that leverages pre-trained Vision-Language Models (VLMs) to bridge temporal, visual, and textual modalities for enhanced forecasting. Our framework comprises three key components: (1) a Retrieval-Augmented Learner, which extracts enriched temporal features through memory bank interactions; (2) a Vision-Augmented Learner, which encodes time series as informative images; and (3) a Text-Augmented Learner, which generates contextual textual descriptions. These components collaborate with frozen pre-trained VLMs to produce multimodal embeddings, which are then fused with temporal features for final prediction. Extensive experiments across diverse datasets demonstrate that Time-VLM achieves superior performance, particularly in few-shot and zero-shot scenarios, thereby establishing a new direction for multimodal time series forecasting.

Position: Empowering Time Series Reasoning with Multimodal LLMs

Understanding time series data is crucial for multiple real-world applications. While large language models (LLMs) show promise in time series tasks, current approaches often rely on numerical data alone, overlooking the multimodal nature of time-dependent information, such as textual descriptions, visual data, and audio signals. Moreover, these methods underutilize LLMs' reasoning capabilities, limiting the analysis to surface-level interpretations instead of deeper temporal and multimodal reasoning. In this position paper, we argue that multimodal LLMs (MLLMs) can enable more powerful and flexible reasoning for time series analysis, enhancing decision-making and real-world applications. We call on researchers and practitioners to leverage this potential by developing strategies that prioritize trust, interpretability, and robust reasoning in MLLMs. Lastly, we highlight key research directions, including novel reasoning paradigms, architectural innovations, and domain-specific applications, to advance time series reasoning with MLLMs.

Detecting Zero-Day Attacks in Digital Substations via In-Context Learning

The occurrences of cyber attacks on the power grids have been increasing every year, with novel attack techniques emerging every year. In this paper, we address the critical challenge of detecting novel/zero-day attacks in digital substations that employ the IEC-61850 communication protocol. While many heuristic and machine learning (ML)-based methods have been proposed for attack detection in IEC-61850 digital substations, generalization to novel or zero-day attacks remains challenging. We propose an approach that leverages the in-context learning (ICL) capability of the transformer architecture, the fundamental building block of large language models. The ICL approach enables the model to detect zero-day attacks and learn from a few examples of that attack without explicit retraining. Our experiments on the IEC-61850 dataset demonstrate that the proposed method achieves more than $85\%$ detection accuracy on zero-day attacks while the existing state-of-the-art baselines fail. This work paves the way for building more secure and resilient digital substations of the future.

Rethinking Hardware Impairments in Multi-User Systems: Can FAS Make a Difference?

In this paper, we analyze the role of fluid antenna systems (FAS) in multi-user systems with hardware impairments (HIs). Specifically, we investigate a scenario where a base station (BS) equipped with multiple fluid antennas communicates with multiple users (CUs), each equipped with a single fluid antenna. Our objective is to maximize the minimum communication rate among all users by jointly optimizing the BS's transmit beamforming, the positions of its transmit fluid antennas, and the positions of the CUs' receive fluid antennas. To address this non-convex problem, we propose a block coordinate descent (BCD) algorithm integrating semidefinite relaxation (SDR), rank-one constraint relaxation (SRCR), successive convex approximation (SCA), and majorization-minimization (MM). Simulation results demonstrate that FAS significantly enhances system performance and robustness, with notable gains when both the BS and CUs are equipped with fluid antennas. Even under low transmit power conditions, deploying FAS at the BS alone yields substantial performance gains. However, the effectiveness of FAS depends on the availability of sufficient movement space, as space constraints may limit its benefits compared to fixed antenna strategies. Our findings highlight the potential of FAS to mitigate HIs and enhance multi-user system performance, while emphasizing the need for practical deployment considerations.

Boosting Alignment for Post-Unlearning Text-to-Image Generative Models

Large-scale generative models have shown impressive image-generation capabilities, propelled by massive data. However, this often inadvertently leads to the generation of harmful or inappropriate content and raises copyright concerns. Driven by these concerns, machine unlearning has become crucial to effectively purge undesirable knowledge from models. While existing literature has studied various unlearning techniques, these often suffer from either poor unlearning quality or degradation in text-image alignment after unlearning, due to the competitive nature of these objectives. To address these challenges, we propose a framework that seeks an optimal model update at each unlearning iteration, ensuring monotonic improvement on both objectives. We further derive the characterization of such an update. In addition, we design procedures to strategically diversify the unlearning and remaining datasets to boost performance improvement. Our evaluation demonstrates that our method effectively removes target classes from recent diffusion-based generative models and concepts from stable diffusion models while maintaining close alignment with the models' original trained states, thus outperforming state-of-the-art baselines. Our code will be made available at \url{https://github.com/reds-lab/Restricted_gradient_diversity_unlearning.git}.

Augmenting Minds or Automating Skills: The Differential Role of Human Capital in Generative AI's Impact on Creative Tasks

Generative AI is rapidly reshaping creative work, raising critical questions about its beneficiaries and societal implications. This study challenges prevailing assumptions by exploring how generative AI interacts with diverse forms of human capital in creative tasks. Through two random controlled experiments in flash fiction writing and song composition, we uncover a paradox: while AI democratizes access to creative tools, it simultaneously amplifies cognitive inequalities. Our findings reveal that AI enhances general human capital (cognitive abilities and education) by facilitating adaptability and idea integration but diminishes the value of domain-specific expertise. We introduce a novel theoretical framework that merges human capital theory with the automation-augmentation perspective, offering a nuanced understanding of human-AI collaboration. This framework elucidates how AI shifts the locus of creative advantage from specialized expertise to broader cognitive adaptability. Contrary to the notion of AI as a universal equalizer, our work highlights its potential to exacerbate disparities in skill valuation, reshaping workplace hierarchies and redefining the nature of creativity in the AI era. These insights advance theories of human capital and automation while providing actionable guidance for organizations navigating AI integration amidst workforce inequalities.