Value function initialization (VFI) is an effective way to achieve a jumpstart in reinforcement learning (RL) by leveraging value estimates from prior tasks. While this approach is well established in tabular settings, extending it to deep reinforcement learning (DRL) poses challenges due to the continuous nature of the state-action space, the noisy approximations of neural networks, and the impracticality of storing all past models for reuse. In this work, we address these challenges and introduce DQInit, a method that adapts value function initialization to DRL. DQInit reuses compact tabular Q-values extracted from previously solved tasks as a transferable knowledge base. It employs a knownness-based mechanism to softly integrate these transferred values into underexplored regions and gradually shift toward the agent's learned estimates, avoiding the limitations of fixed time decay. Our approach offers a novel perspective on knowledge transfer in DRL by relying solely on value estimates rather than policies or demonstrations, effectively combining the strengths of jumpstart RL and policy distillation while mitigating their drawbacks. Experiments across multiple continuous control tasks demonstrate that DQInit consistently improves early learning efficiency, stability, and overall performance compared to standard initialization and existing transfer techniques.

Federated learning (FL) provides a decentralized framework that enables universal model training through collaborative efforts on mobile nodes, such as smart vehicles in the Internet of Vehicles (IoV). Each smart vehicle acts as a mobile client, contributing to the process without uploading local data. This method leverages non-independent and identically distributed (non-IID) training data from different vehicles, influenced by various driving patterns and environmental conditions, which can significantly impact model convergence and accuracy. Although client selection can be a feasible solution for non-IID issues, it faces challenges related to selection metrics. Traditional metrics evaluate client data quality independently per round and require client selection after all clients complete local training, leading to resource wastage from unused training results. In the IoV context, where vehicles have limited connectivity and computational resources, information asymmetry in client selection risks clients submitting false information, potentially making the selection ineffective. To tackle these challenges, we propose a novel Long-term Client-Selection Federated Learning based on Truthful Auction (LCSFLA). This scheme maximizes social welfare with consideration of long-term data quality using a new assessment mechanism and energy costs, and the advised auction mechanism with a deposit requirement incentivizes client participation and ensures information truthfulness. We theoretically prove the incentive compatibility and individual rationality of the advised incentive mechanism. Experimental results on various datasets, including those from IoV scenarios, demonstrate its effectiveness in mitigating performance degradation caused by non-IID data.

Self-evaluation, a model's ability to assess the correctness of its own output, is crucial for Large Multimodal Models (LMMs) to achieve self-improvement in multi-turn conversations, yet largely absent in foundation models. Recent work has employed reinforcement learning (RL) to enhance self-evaluation; however, its fixed reward mechanism suffers from reward hacking when optimizing multiple training objectives, leading to model collapse. In this paper we propose AdaPO, an online reinforcement learning framework capable of adaptively adjusting training objective in real time according to the current training state for each task. Specifically, to mitigate reward hacking , AdaPO introduces an Adaptive Reward Model (ARM) and a Reward Aware Dynamic KL Regularization mechanism. ARM assesses the task's training state from the distribution of model generated multi-turn trajectories' performance. Reward Aware Dynamic KL replaces a fixed penalty with dynamic coefficients which is modulated by the reward gap between different multi-turn situations. Notably, our method automatically and smoothly adjusts its learning focus based on sub-tasks' training progress without manual intervention. Extensive experiments over 8 benchmarks and various models show that our method significantly enhances both direct reasoning and self-evaluation capability. We will release our code to contribute to the community.

Abrupt resolution changes in virtual reality (VR) streaming can significantly impair the quality-of-experience (QoE) of users, particularly during transitions from high to low resolutions. Existing QoE models and transmission schemes inadequately address the perceptual impact of these shifts. To bridge this gap, this article proposes, for the first time, an innovative physiological signal-driven QoE modeling and optimization framework that fully leverages users' electroencephalogram (EEG), electrocardiogram (ECG), and skin activity signals. Integrated the proposed QoE framework into the radio access network (RAN) via a deep reinforcement learning (DRL) framework, adaptive transmission strategies have been implemented to allocate radio resources dynamically, which mitigates short-term channel fluctuations and adjusts frame resolution in response to channel variations caused by user mobility. By prioritizing long-term resolution while minimizing abrupt transitions, the proposed solution achieves an 88.7% improvement in resolution and an 81.0% Experimental results demonstrate the effectiveness of this physiological signal-driven strategy, underscoring the promise of edge AI in immersive media services. While this technology enables unprecedented engagement in applications ranging from event viewing to interactive education, its reliance on wireless transmission poses critical challenges. The uncompressed data rates exceeding 1 Gbps and latency thresholds below 20 ms impose stringent demands on network infrastructure, particularly in mobile scenarios where channel fluctuations and user mobility degrade service consistency. Traditional quality of service (QoS) metrics (e.g., bandwidth, jitter, and packet loss) provide necessary but insufficient insights into user satisfaction, necessitating perceptual quality of experience (QoE) frameworks tailored to user's unique requirements [3]-[5].

We present Aryabhata 1.0, a compact 7B parameter math reasoning model optimized for the Indian academic exam, the Joint Entrance Examination (JEE). Despite rapid progress in large language models (LLMs), current models often remain unsuitable for educational use. Aryabhata 1.0 is built by merging strong open-weight reasoning models, followed by supervised fine-tuning (SFT) with curriculum learning on verified chain-of-thought (CoT) traces curated through best-of-$n$ rejection sampling. To further boost performance, we apply reinforcement learning with verifiable rewards (RLVR) using A2C objective with group-relative advantage estimation along with novel exploration strategies such as Adaptive Group Resizing and Temperature Scaling. Evaluated on both in-distribution (JEE Main 2025) and out-of-distribution (MATH, GSM8K) benchmarks, Aryabhata outperforms existing models in accuracy and efficiency, while offering pedagogically useful step-by-step reasoning. We release Aryabhata as a foundation model to advance exam-centric, open-source small language models. This marks our first open release for community feedback (https://huggingface.co/PhysicsWallahAI/Aryabhata-1.0); PW is actively training future models to further improve learning outcomes for students.

As large language models (LLMs) grow more capable, they face growing vulnerability to sophisticated jailbreak attacks. While developers invest heavily in alignment finetuning and safety guardrails, researchers continue publishing novel attacks, driving progress through adversarial iteration. This dynamic mirrors a strategic game of continual evolution. However, two major challenges hinder jailbreak development: the high cost of querying top-tier LLMs and the short lifespan of effective attacks due to frequent safety updates. These factors limit cost-efficiency and practical impact of research in jailbreak attacks. To address this, we propose MetaCipher, a low-cost, multi-agent jailbreak framework that generalizes across LLMs with varying safety measures. Using reinforcement learning, MetaCipher is modular and adaptive, supporting extensibility to future strategies. Within as few as 10 queries, MetaCipher achieves state-of-the-art attack success rates on recent malicious prompt benchmarks, outperforming prior jailbreak methods. We conduct a large-scale empirical evaluation across diverse victim models and benchmarks, demonstrating its robustness and adaptability. Warning: This paper contains model outputs that may be offensive or harmful, shown solely to demonstrate jailbreak efficacy.

AI-generated text is proliferating across domains, from creative writing and journalism to marketing content and scientific articles. Models can follow user-provided instructions to generate coherent and grammatically correct outputs but in this work, we study a more fundamental question: how do we evaluate and improve the writing quality of AI-generated text? Writing quality assessment has received less attention from the community, in part because it is fundamentally subjective and requires expertise. We first introduce the Writing Quality Benchmark (WQ) by consolidating five writing-preference datasets into 4,729 writing quality judgments. Our experiments show that most of the competitive baselines, including state-of-the-art LLMs that excel at reasoning tasks, barely outperform random baselines on WQ. We then train specialized Writing Quality Reward Models (WQRM) of various sizes for writing quality assessment that demonstrate strong generalization on four out-of-distribution test sets and 74% accuracy on the WQ benchmark. To further show WQRM's practical benefits during inference, we leverage additional test-time compute to generate and rank multiple candidate revisions, allowing us to select higher-quality outputs from an initial draft. Human evaluation with 9 experienced writers confirm that WQRM-based selection produces writing samples preferred by experts 66% overall, and 72.2% when the reward gap is larger than 1 point. We release our datasets and models to encourage community engagement with writing quality assessment and development of AI writing systems better aligned with human preferences.

Sign language, which contains hand movements, facial expressions and bodily gestures, is a significant medium for communicating with hard-of-hearing people. A well-trained sign language community communicates easily, but those who don't know sign language face significant challenges. Recognition and generation are basic communication methods between hearing and hard-of-hearing individuals. Despite progress in recognition, sign language generation still needs to be explored. The Progressive Growing of Generative Adversarial Network (ProGAN) excels at producing high-quality images, while the Self-Attention Generative Adversarial Network (SAGAN) generates feature-rich images at medium resolutions. Balancing resolution and detail is crucial for sign language image generation. We are developing a Generative Adversarial Network (GAN) variant that combines both models to generate feature-rich, high-resolution, and class-conditional sign language images. Our modified Attention-based model generates high-quality images of Indian Sign Language letters, numbers, and words, outperforming the traditional ProGAN in Inception Score (IS) and Fréchet Inception Distance (FID), with improvements of 3.2 and 30.12, respectively. Additionally, we are publishing a large dataset incorporating high-quality images of Indian Sign Language alphabets, numbers, and 129 words.

Recent advances in Vision-Language Models (VLMs) have demonstrated impressive capabilities in perception and reasoning. However, the ability to perform causal inference -- a core aspect of human cognition -- remains underexplored, particularly in multimodal settings. In this study, we introduce InfoCausalQA, a novel benchmark designed to evaluate causal reasoning grounded in infographics that combine structured visual data with textual context. The benchmark comprises two tasks: Task 1 focuses on quantitative causal reasoning based on inferred numerical trends, while Task 2 targets semantic causal reasoning involving five types of causal relations: cause, effect, intervention, counterfactual, and temporal. We manually collected 494 infographic-text pairs from four public sources and used GPT-4o to generate 1,482 high-quality multiple-choice QA pairs. These questions were then carefully revised by humans to ensure they cannot be answered based on surface-level cues alone but instead require genuine visual grounding. Our experimental results reveal that current VLMs exhibit limited capability in computational reasoning and even more pronounced limitations in semantic causal reasoning. Their significantly lower performance compared to humans indicates a substantial gap in leveraging infographic-based information for causal inference. Through InfoCausalQA, we highlight the need for advancing the causal reasoning abilities of multimodal AI systems.

Principle Engraving: In the third stage, we fine-tune the original LLM (the base model) on the self-aligned responses, generated by the LLM itself through prompting, while pruning the principles and demonstrations for the fine-tuned model.