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.

Sequence and channel mixers, the core mechanism in sequence models, have become the de facto standard in time series analysis (TSA). However, recent studies have questioned the necessity of complex sequence mixers, such as attention mechanisms, demonstrating that simpler architectures can achieve comparable or even superior performance. This suggests that the benefits attributed to complex sequencemixers might instead emerge from other architectural or optimization factors. Based on this observation, we pose a central question: Are common sequence mixers necessary for time-series analysis? Therefore, we propose JustDense, an empirical study that systematically replaces sequence mixers in various well-established TSA models with dense layers. Grounded in the MatrixMixer framework, JustDense treats any sequence mixer as a mixing matrix and replaces it with a dense layer. This substitution isolates the mixing operation, enabling a clear theoretical foundation for understanding its role. Therefore, we conducted extensive experiments on 29 benchmarks covering five representative TSA tasks using seven state-of-the-art TSA models to address our research question. The results show that replacing sequence mixers with dense layers yields comparable or even superior performance. In the cases where dedicated sequence mixers still offer benefits, JustDense challenges the assumption that "deeper and more complex architectures are inherently better" in TSA.

With the advent of 5G networks and technologies, ensuring the integrity and performance of packet core traffic is paramount. During network analysis, test files such as Packet Capture (PCAP) files and log files will contain errors if present in the system that must be resolved for better overall network performance, such as connectivity strength and handover quality. Current methods require numerous person-hours to sort out testing results and find the faults. This paper presents a novel AI/ML-driven Fault Analysis (FA) Engine designed to classify successful and faulty frames in PCAP files, specifically within the 5G packet core. The FA engine analyses network traffic using natural language processing techniques to identify anomalies and inefficiencies, significantly reducing the effort time required and increasing efficiency. The FA Engine also suggests steps to fix the issue using Generative AI via a Large Language Model (LLM) trained on several 5G packet core documents. The engine explains the details of the error from the domain perspective using documents such as the 3GPP standards and user documents regarding the internal conditions of the tests. Test results on the ML models show high classification accuracy on the test dataset when trained with 80-20 splits for the successful and failed PCAP files. Future scopes include extending the AI engine to incorporate 4G network traffic and other forms of network data, such as log text files and multimodal systems.

Recent advancements in Large Language Models (LLMs) have revolutionized artificial intelligence, yet developing an effective foundational LLM that balances high performance with computational efficiency remains challenging, especially for emerging research groups. To address this gap, we introduce Motif-2.6B, a 2.6-billion-parameter foundation model designed to democratize advanced LLM capabilities. Motif-2.6B incorporates several innovative architectural enhancements, including Differential Attention and PolyNorm activation functions, which improve long-context comprehension, reduce hallucination, and enhance in-context learning capabilities. We rigorously tested multiple novel architectural components through extensive experimentation to determine the optimal architecture for Motif-2.6B. Comprehensive evaluations demonstrate that Motif-2.6B consistently meets or exceeds the performance of similarly sized state-of-the-art models across diverse benchmarks, showcasing its effectiveness, scalability, and real-world applicability. Through detailed experiments and tailored techniques, Motif-2.6B significantly advances the landscape of efficient, scalable, and powerful foundational LLMs, offering valuable insights and a robust foundation for future research and deployment.

Estimated time of arrival (ETA) for airborne aircraft in real-time is crucial for arrival management in aviation, particularly for runway sequencing. Given the rapidly changing airspace context, the ETA prediction efficiency is as important as its accuracy in a real-time arrival aircraft management system. In this study, we utilize a feature tokenization-based Transformer model to efficiently predict aircraft ETA. Feature tokenization projects raw inputs to latent spaces, while the multi-head self-attention mechanism in the Transformer captures important aspects of the projections, alleviating the need for complex feature engineering. Moreover, the Transformer's parallel computation capability allows it to handle ETA requests at a high frequency, i.e., 1HZ, which is essential for a real-time arrival management system. The model inputs include raw data, such as aircraft latitude, longitude, ground speed, theta degree for the airport, day and hour from track data, the weather context, and aircraft wake turbulence category. With a data sampling rate of 1HZ, the ETA prediction is updated every second. We apply the proposed aircraft ETA prediction approach to Singapore Changi Airport (ICAO Code: WSSS) using one-month Automatic Dependent Surveillance-Broadcast (ADS-B) data from October 1 to October 31, 2022. In the experimental evaluation, the ETA modeling covers all aircraft within a range of 10NM to 300NM from WSSS. The results show that our proposed method method outperforms the commonly used boosting tree based model, improving accuracy by 7\% compared to XGBoost, while requiring only 39\% of its computing time. Experimental results also indicate that, with 40 aircraft in the airspace at a given timestamp, the ETA inference time is only 51.7 microseconds, making it promising for real-time arrival management systems.

--Radio map (RM) has recently attracted much attention since it can provide real-time and accurate spatial channel information for 6G services and applications. However, current deep learning-based methods for RM construction exhibit well known accuracy-efficiency trade-off. In this paper, we introduce RadioMamba, a hybrid Mamba-UNet architecture for RM construction to address the trade-off. Generally, accurate RM construction requires modeling long-range spatial dependencies, reflecting the global nature of wave propagation physics. This hybrid design generates feature representations that capture both global context and local detail. Experiments show that RadioMamba achieves higher accuracy than existing methods, including diffusion models, while operating nearly 20 times faster and using only 2.9% of the model parameters. By improving both accuracy and efficiency, RadioMamba presents a viable approach for real-time intelligent optimization in next generation wireless systems. The continuous advancement towards sixth-generation (6G) wireless networks is enabling a future with the internet of things (IoT), autonomous systems, and immersive cyber-physical experiences [1], [2]. A key component of this evolution is the network digital twin (NDT) [4], [5], a high-fidelity virtual replica of the physical network environment that enables simulation, prediction, and optimization in real-time [6]-[8]. This work was supported by the National Key Research and Development Program of China (2024YFB2907500). Honggang Jia, Nan Cheng, Xiucheng Wang, Conghao Zhou, Ruijin Sun are with the State Key Laboratory of ISN and School of Telecommunications Engineering, Xidian University, Xi'an 710071, China (e-mail: ji-ahg@stu.xidian.edu.cn; Nan Cheng is the corresponding author . Xuemin (Sherman) Shen is with the Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, N2L 3G1, Canada (e-mail: sshen@uwaterloo.ca).

Text-guided color editing in images and videos is a fundamental yet unsolved problem, requiring fine-grained manipulation of color attributes, including albedo, light source color, and ambient lighting, while preserving physical consistency in geometry, material properties, and light-matter interactions. Existing training-free methods offer broad applicability across editing tasks but struggle with precise color control and often introduce visual inconsistency in both edited and non-edited regions. In this work, we present ColorCtrl, a training-free color editing method that leverages the attention mechanisms of modern Multi-Modal Diffusion Transformers (MM-DiT). By disentangling structure and color through targeted manipulation of attention maps and value tokens, our method enables accurate and consistent color editing, along with word-level control of attribute intensity. Our method modifies only the intended regions specified by the prompt, leaving unrelated areas untouched. Extensive experiments on both SD3 and FLUX.1-dev demonstrate that ColorCtrl outperforms existing training-free approaches and achieves state-of-the-art performances in both edit quality and consistency. Furthermore, our method surpasses strong commercial models such as FLUX.1 Kontext Max and GPT-4o Image Generation in terms of consistency. When extended to video models like CogVideoX, our approach exhibits greater advantages, particularly in maintaining temporal coherence and editing stability. Finally, our method also generalizes to instruction-based editing diffusion models such as Step1X-Edit and FLUX.1 Kontext dev, further demonstrating its versatility.

Retrieval-Augmented Generation (RAG) and its Multimodal Retrieval-Augmented Generation (MRAG) significantly improve the knowledge coverage and contextual understanding of Large Language Models (LLMs) by introducing external knowledge sources. However, retrieval and multimodal fusion obscure content provenance, rendering existing membership inference methods unable to reliably attribute generated outputs to pre-training, external retrieval, or user input, thus undermining privacy leakage accountability To address these challenges, we propose the first Source-aware Membership Audit (SMA) that enables fine-grained source attribution of generated content in a semi-black-box setting with retrieval control capabilities. To address the environmental constraints of semi-black-box auditing, we further design an attribution estimation mechanism based on zero-order optimization, which robustly approximates the true influence of input tokens on the output through large-scale perturbation sampling and ridge regression modeling. In addition, SMA introduces a cross-modal attribution technique that projects image inputs into textual descriptions via MLLMs, enabling token-level attribution in the text modality, which for the first time facilitates membership inference on image retrieval traces in MRAG systems. This work shifts the focus of membership inference from 'whether the data has been memorized' to 'where the content is sourced from', offering a novel perspective for auditing data provenance in complex generative systems.

Modeling multi-interests has arisen as a core problem in real-world RS. Current multi-interest retrieval methods pose three major challenges: 1) Interests, typically extracted from predefined external knowledge, are invariant. Failed to dynamically evolve with users' real-time consumption preferences. 2) Online inference typically employs an over-exploited strategy, mainly matching users' existing interests, lacking proactive exploration and discovery of novel and long-tail interests. To address these challenges, we propose a novel retrieval framework named SPARC(Soft Probabilistic Adaptive Retrieval Model via Codebooks). Our contribution is two folds. First, the framework utilizes Residual Quantized Variational Autoencoder (RQ-VAE) to construct a discretized interest space. It achieves joint training of the RQ-VAE with the industrial large scale recommendation model, mining behavior-aware interests that can perceive user feedback and evolve dynamically. Secondly, a probabilistic interest module that predicts the probability distribution over the entire dynamic and discrete interest space. This facilitates an efficient "soft-search" strategy during online inference, revolutionizing the retrieval paradigm from "passive matching" to "proactive exploration" and thereby effectively promoting interest discovery. Online A/B tests on an industrial platform with tens of millions daily active users, have achieved substantial gains in business metrics: +0.9% increase in user view duration, +0.4% increase in user page views (PV), and a +22.7% improvement in PV500(new content reaching 500 PVs in 24 hours). Offline evaluations are conducted on open-source Amazon Product datasets. Metrics, such as Recall@K and Normalized Discounted Cumulative Gain@K(NDCG@K), also showed consistent improvement. Both online and offline experiments validate the efficacy and practical value of the proposed method.

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.