Large language models (LLMs) have significantly improved their reasoning capabilities; however, they still struggle with complex multi-step mathematical problem-solving due to error propagation, lack of self-correction, and limited adaptability to diverse reasoning styles. Existing methods rely on static fine-tuning or prompt engineering, which fail to generalize across problem complexities, while the scarcity of high-quality preference data further hinders reliable reasoning. We introduce SPHERE, a self-evolving data generation pipeline that enhances reasoning in small language models (SLMs) by iteratively generating, correcting, and diversifying reasoning chains. SPHERE operates in three stages: (i) Self-Generation, where the model autonomously constructs problem-solving steps; (ii) Self-Correction, enabling it to identify and rectify errors; and (iii) Diversity Induction, improving robustness through multiple valid reasoning trajectories. This self-evolution mechanism strengthens mathematical reasoning and enhances model reliability. Evaluations on MATH 500, GSM8K, AIME, AMC, and Olympiad show that SPHERE-trained models achieve significant gains over their base versions and match/surpass GPT-4o on certain benchmarks. Our findings demonstrate that self-evolving models can close the reasoning gap between SLMs and state-of-the-art LLMs, making mathematical AI more reliable, scalable, and efficient.

The reliability of large language models remains a critical challenge, particularly due to their susceptibility to hallucinations and factual inaccuracies during text generation. Existing solutions either underutilize models' self-correction with preemptive strategies or use costly post-hoc verification. To further explore the potential of real-time self-verification and correction, we present Dynamic Self-Verify Decoding (DSVD), a novel decoding framework that enhances generation reliability through real-time hallucination detection and efficient error correction. DSVD integrates two key components: (1) parallel self-verification architecture for continuous quality assessment, (2) dynamic rollback mechanism for targeted error recovery. Extensive experiments across five benchmarks demonstrate DSVD's effectiveness, achieving significant improvement in truthfulness (Quesetion-Answering) and factual accuracy (FActScore). Results show the DSVD can be further incorporated with existing faithful decoding methods to achieve stronger performance. Our work establishes that real-time self-verification during generation offers a viable path toward more trustworthy language models without sacrificing practical deployability.

Deploying language models often requires handling model size vs. performance trade-offs to satisfy downstream latency constraints while preserving the model's usefulness. Model distillation is commonly employed to reduce model size while maintaining acceptable performance. However, distillation can be inefficient since it involves multiple training steps. In this work, we introduce MODULARSTARENCODER, a modular multi-exit encoder with 1B parameters, useful for multiple tasks within the scope of code retrieval. MODULARSTARENCODER is trained with a novel self-distillation mechanism that significantly improves lower-layer representations-allowing different portions of the model to be used while still maintaining a good trade-off in terms of performance. Our architecture focuses on enhancing text-to-code and code-to-code search by systematically capturing syntactic and semantic structures across multiple levels of representation. Specific encoder layers are targeted as exit heads, allowing higher layers to guide earlier layers during training. This self-distillation effect improves intermediate representations, increasing retrieval recall at no extra training cost. In addition to the multi-exit scheme, our approach integrates a repository-level contextual loss that maximally utilizes the training context window, further enhancing the learned representations. We also release a new dataset constructed via code translation, seamlessly expanding traditional text-to-code benchmarks with code-to-code pairs across diverse programming languages. Experimental results highlight the benefits of self-distillation through multi-exit supervision.

Graph Neural Networks (GNNs) are a new research frontier with various applications and successes. The end-to-end inference for all nodes, is common for GNN embedding models, which are widely adopted in applications like recommendation and advertising. While sharing opportunities arise in GNN tasks (i.e., inference for a few nodes and training), the potential for sharing in full graph end-to-end inference is largely underutilized because traditional efforts fail to fully extract sharing benefits due to overwhelming overheads or excessive memory usage. This paper introduces Deal, a distributed GNN inference system that is dedicated to end-to-end inference for all nodes for graphs with multi-billion edges. First, we unveil and exploit an untapped sharing opportunity during sampling, and maximize the benefits from sharing during subsequent GNN computation. Second, we introduce memory-saving and communication-efficient distributed primitives for lightweight 1-D graph and feature tensor collaborative partitioning-based distributed inference. Third, we introduce partitioned, pipelined communication and fusing feature preparation with the first GNN primitive for end-to-end inference. With Deal, the end-to-end inference time on real-world benchmark datasets is reduced up to 7.70 x and the graph construction time is reduced up to 21.05 x, compared to the state-of-the-art.

Image quality plays an important role in the performance of deep neural networks (DNNs) and DNNs have been widely shown to exhibit sensitivity to changes in imaging conditions. Large-scale datasets often contain images under a wide range of conditions prompting a need to quantify and understand their underlying quality distribution in order to better characterize DNN performance and robustness. Aligning the sensitivities of image quality metrics and DNNs ensures that estimates of quality can act as proxies for image/dataset difficulty independent of the task models trained/evaluated on the data. Conventional image quality assessment (IQA) seeks to measure and align quality relative to human perceptual judgments, but here we seek a quality measure that is not only sensitive to imaging conditions but also well-aligned with DNN sensitivities. We first ask whether conventional IQA metrics are also informative of DNN performance. In order to answer this question, we reframe IQA from a causal perspective and examine conditions under which quality metrics are predictive of DNN performance. We show theoretically and empirically that current IQA metrics are weak predictors of DNN performance in the context of classification. We then use our causal framework to provide an alternative formulation and a new image quality metric that is more strongly correlated with DNN performance and can act as a prior on performance without training new task models. Our approach provides a means to directly estimate the quality distribution of large-scale image datasets towards characterizing the relationship between dataset composition and DNN performance. Ensuring the robustness of deep neural networks (DNNs) to real-world imaging conditions is crucial for safety-and cost-critical applications.

Cellular automata (CA) models are widely used to simulate complex systems with emergent behaviors, but identifying hidden parameters that govern their dynamics remains a significant challenge. This study explores the use of Convolutional Neural Networks (CNN) to identify jump parameters in a two-dimensional CA model. We propose a custom CNN architecture trained on CA-generated data to classify jump parameters, which dictates the neighborhood size and movement rules of cells within the CA. Experiments were conducted across varying domain sizes (25 x 25 to 150 x 150) and CA iterations (0 to 50), demonstrating that the accuracy improves with larger domain sizes, as they provide more spatial information for parameter estimation. Interestingly, while initial CA iterations enhance the performance, increasing the number of iterations beyond a certain threshold does not significantly improve accuracy, suggesting that only specific temporal information is relevant for parameter identification. The proposed CNN achieves competitive accuracy (89.31) compared to established architectures like LeNet-5 and AlexNet, while offering significantly faster inference times, making it suitable for real-time applications. This study highlights the potential of CNNs as a powerful tool for fast and accurate parameter estimation in CA models, paving the way for their use in more complex systems and higher-dimensional domains. Future work will explore the identification of multiple hidden parameters and extend the approach to three-dimensional CA models.

The COVID-19 pandemic has profoundly affected the normal course of life -- from lock-downs and virtual meetings to the unprecedentedly swift creation of vaccines. To halt the COVID-19 pandemic, the world has started preparing for the global vaccine roll-out. In an effort to navigate the immense volume of information about COVID-19, the public has turned to social networks. Among them, $\mathbb{X}$ (formerly Twitter) has played a key role in distributing related information. Most people are not trained to interpret medical research and remain skeptical about the efficacy of new vaccines. Measuring their reactions and perceptions is gaining significance in the fight against COVID-19. To assess the public perception regarding the COVID-19 vaccine, our work applies a sentiment analysis approach, using natural language processing of $\mathbb{X}$ data. We show how to use textual analytics and textual data visualization to discover early insights (for example, by analyzing the most frequently used keywords and hashtags). Furthermore, we look at how people's sentiments vary across the countries. Our results indicate that although the overall reaction to the vaccine is positive, there are also negative sentiments associated with the tweets, especially when examined at the country level. Additionally, from the extracted tweets, we manually labeled 100 tweets as positive and 100 tweets as negative and trained various One-Class Classifiers (OCCs). The experimental results indicate that the S-SVDD classifiers outperform other OCCs.

The nnU-Net framework has played a crucial role in medical image segmentation and has become the gold standard in multitudes of applications targeting different diseases, organs, and modalities. However, so far it has been used primarily in a centralized approach where the data collected from hospitals are stored in one center and used to train the nnU-Net. This centralized approach has various limitations, such as leakage of sensitive patient information and violation of patient privacy. Federated learning is one of the approaches to train a segmentation model in a decentralized manner that helps preserve patient privacy. In this paper, we propose FednnU-Net, a federated learning extension of nnU-Net. We introduce two novel federated learning methods to the nnU-Net framework - Federated Fingerprint Extraction (FFE) and Asymmetric Federated Averaging (AsymFedAvg) - and experimentally show their consistent performance for breast, cardiac and fetal segmentation using 6 datasets representing samples from 18 institutions. Additionally, to further promote research and deployment of decentralized training in privacy constrained institutions, we make our plug-n-play framework public. The source-code is available at https://github.com/faildeny/FednnUNet .

In the realm of Intelligent Tutoring System (ITS), the accurate assessment of students' knowledge states through Knowledge Tracing (KT) is crucial for personalized learning. However, due to data bias, $\textit{i.e.}$, the unbalanced distribution of question groups ($\textit{e.g.}$, concepts), conventional KT models are plagued by cognitive bias, which tends to result in cognitive underload for overperformers and cognitive overload for underperformers. More seriously, this bias is amplified with the exercise recommendations by ITS. After delving into the causal relations in the KT models, we identify the main cause as the confounder effect of students' historical correct rate distribution over question groups on the student representation and prediction score. Towards this end, we propose a Disentangled Knowledge Tracing (DisKT) model, which separately models students' familiar and unfamiliar abilities based on causal effects and eliminates the impact of the confounder in student representation within the model. Additionally, to shield the contradictory psychology ($\textit{e.g.}$, guessing and mistaking) in the students' biased data, DisKT introduces a contradiction attention mechanism. Furthermore, DisKT enhances the interpretability of the model predictions by integrating a variant of Item Response Theory. Experimental results on 11 benchmarks and 3 synthesized datasets with different bias strengths demonstrate that DisKT significantly alleviates cognitive bias and outperforms 16 baselines in evaluation accuracy.

Prior work has demonstrated a consistent tendency in neural networks engaged in continual learning tasks, wherein intermediate task similarity results in the highest levels of catastrophic interference. This phenomenon is attributed to the network's tendency to reuse learned features across tasks. However, this explanation heavily relies on the premise that neuron specialisation occurs, i.e. the emergence of localised representations. Our investigation challenges the validity of this assumption. Using theoretical frameworks for the analysis of neural networks, we show a strong dependence of specialisation on the initial condition. More precisely, we show that weight imbalance and high weight entropy can favour specialised solutions. We then apply these insights in the context of continual learning, first showing the emergence of a monotonic relation between task-similarity and forgetting in non-specialised networks. {Finally, we show that specialization by weight imbalance is beneficial on the commonly employed elastic weight consolidation regularisation technique.