Xiaomai is an intelligent tutoring system (ITS) designed to help Chinese college students in learning advanced mathematics and preparing for the graduate school math entrance exam. This study investigates two distinctive features within Xiaomai: the incorporation of free-response questions with automatic feedback and the metacognitive element of reflecting on self-made errors. An experiment was conducted to evaluate the impact of these features on mathematics learning. One hundred and twenty college students were recruited and randomly assigned to four conditions: (1) multiple-choice questions without reflection, (2) multiple-choice questions with reflection, (3) free-response questions without reflection, and (4) free-response questions with reflection. Students in the multiple-choice conditions demonstrated better practice performance and learning outcomes compared to their counterparts in the freeresponse conditions. Additionally, the incorporation of error reflection did not yield a significant impact on students' practice performance or learning outcomes. These findings indicate that current design of free-response questions and the metacognitive feature of error reflection do not enhance the efficacy of the math ITS. This study highlights the need for redesign or enhancement of Xiaomai to optimize its effectiveness in facilitating advanced mathematics learning.

Solving algebra problems (APs) continues to attract significant research interest as evidenced by the large number of algorithms and theories proposed over the past decade. Despite these important research contributions, however, the body of work remains incomplete in terms of theoretical justification and scope. The current contribution intends to fill the gap by developing a review framework that aims to lay a theoretical base, create an evaluation scheme, and extend the scope of the investigation. This paper first develops the State Transform Theory (STT), which emphasizes that the problem-solving algorithms are structured according to states and transforms unlike the understanding that underlies traditional surveys which merely emphasize the progress of transforms. The STT, thus, lays the theoretical basis for a new framework for reviewing algorithms. This new construct accommodates the relation-centric algorithms for solving both word and diagrammatic algebra problems. The latter not only highlights the necessity of introducing new states but also allows revelation of contributions of individual algorithms obscured in prior reviews without this approach.

Recently, slow-thinking reasoning systems, such as o1, have demonstrated remarkable capabilities in solving complex reasoning tasks. These systems typically engage in an extended thinking process before responding to a query, allowing them to generate more thorough, accurate, and well-reasoned solutions. These systems are primarily developed and maintained by industry, with their core techniques not publicly disclosed. In response, an increasing number of studies from the research community aim to explore the technical foundations underlying these powerful reasoning systems. Building on these prior efforts, this paper presents a reproduction report on implementing o1-like reasoning systems. We introduce an ``imitate, explore, and self-improve'' framework, denoted as \textbf{STILL-2}, as our primary technical approach to train the reasoning model. In the initial phase, we use distilled long-form thought data to fine-tune the reasoning model, enabling it to invoke a slow-thinking mode. The model is then encouraged to explore challenging problems by generating multiple rollouts, which can result in increasingly more high-quality trajectories that lead to correct answers. Furthermore, the model undergoes self-improvement by iteratively refining its training dataset. To verify the effectiveness of this approach, we conduct extensive experiments on three challenging benchmarks. The experimental results demonstrate that our approach achieves competitive performance compared to industry-level reasoning systems on these benchmarks.

Knowing others' intentions and taking others' perspectives are two core components of human intelligence typically considered as instantiations of theory of mind. Infiltrating machines with these abilities is an important step towards building human-level artificial intelligence. We here investigate intentionality understanding and perspective-taking in Vision Language Models and, for the purpose, we have created IntentBench and PerspectBench datasets, which contain over 400 cognitive experiments grounded in real-world scenarios and classic cognitive tasks. Surprisingly, we find that VLMs achieve high performance in intentionality understanding but lower performance in perspective-taking using our two datasets. This challenges the common belief in the cognitive science literature that perspective-taking at the corresponding modality is necessary for intentionality understanding.

Large language models (LLMs) can refine their responses based on feedback, enabling self-improvement through iterative training or test-time refinement. However, existing methods predominantly focus on refinement within the same reasoning format, which may lead to non-correcting behaviors. We propose CaP, a novel approach that uses external tools to refine chain-of-thought (CoT) responses generated by the same or other LLMs. CaP employs a two-stage training process: supervised fine-tuning followed by preference optimization with DPO variants. Our observations highlight the critical role of preference optimization in enabling effective refinement. Additionally, we compare several sampling strategies to leverage CoT and tools at inference time. Experimental results demonstrate CaP's potential for effective cross-reasoning refinement and efficient inference.

This paper introduces mixed-integer optimization methods to solve regression problems that incorporate fairness metrics. We propose an exact formulation for training fair regression models. To tackle this computationally hard problem, we study the polynomially-solvable single-factor and single-observation subproblems as building blocks and derive their closed convex hull descriptions. Strong formulations obtained for the general fair regression problem in this manner are utilized to solve the problem with a branch-and-bound algorithm exactly or as a relaxation to produce fair and accurate models rapidly. Moreover, to handle large-scale instances, we develop a coordinate descent algorithm motivated by the convex-hull representation of the single-factor fair regression problem to improve a given solution efficiently. Numerical experiments conducted on fair least squares and fair logistic regression problems show competitive statistical performance with state-of-the-art methods while significantly reducing training times.

Physical adversarial examples (PAEs) are regarded as "whistle-blowers" of real-world risks in deep-learning applications. However, current PAE generation studies show limited adaptive attacking ability to diverse and varying scenes. The key challenges in generating dynamic PAEs are exploring their patterns under noisy gradient feedback and adapting the attack to agnostic scenario natures. To address the problems, we present DynamicPAE, the first generative framework that enables scene-aware real-time physical attacks beyond static attacks. Specifically, to train the dynamic PAE generator under noisy gradient feedback, we introduce the residual-driven sample trajectory guidance technique, which redefines the training task to break the limited feedback information restriction that leads to the degeneracy problem. Intuitively, it allows the gradient feedback to be passed to the generator through a low-noise auxiliary task, thereby guiding the optimization away from degenerate solutions and facilitating a more comprehensive and stable exploration of feasible PAEs. To adapt the generator to agnostic scenario natures, we introduce the context-aligned scene expectation simulation process, consisting of the conditional-uncertainty-aligned data module and the skewness-aligned objective re-weighting module. The former enhances robustness in the context of incomplete observation by employing a conditional probabilistic model for domain randomization, while the latter facilitates consistent stealth control across different attack targets by automatically reweighting losses based on the skewness indicator. Extensive digital and physical evaluations demonstrate the superior attack performance of DynamicPAE, attaining a 1.95 $\times$ boost (65.55% average AP drop under attack) on representative object detectors (e.g., Yolo-v8) over state-of-the-art static PAE generating methods.

Non-smooth communication-efficient federated optimization is crucial for many machine learning applications, yet remains largely unexplored theoretically. Recent advancements have primarily focused on smooth convex and non-convex regimes, leaving a significant gap in understanding the non-smooth convex setting. Additionally, existing literature often overlooks efficient server-to-worker communication (downlink), focusing primarily on worker-to-server communication (uplink). We consider a setup where uplink costs are negligible and focus on optimizing downlink communication by improving state-of-the-art schemes like EF21-P (arXiv:2209.15218) and MARINA-P (arXiv:2402.06412) in the non-smooth convex setting. We extend the non-smooth convex theory of EF21-P [Anonymous, 2024], originally developed for single-node scenarios, to the distributed setting, and extend MARINA-P to the non-smooth convex setting. For both algorithms, we prove an optimal $O(1/\sqrt{T})$ convergence rate and establish communication complexity bounds matching classical subgradient methods. We provide theoretical guarantees under constant, decreasing, and adaptive (Polyak-type) stepsizes. Our experiments demonstrate that MARINA-P with correlated compressors outperforms other methods in both smooth non-convex and non-smooth convex settings. This work presents the first theoretical results for distributed non-smooth optimization with server-to-worker compression, along with comprehensive analysis for various stepsize schemes.

Edge Intelligence (EI) has been instrumental in delivering real-time, localized services by leveraging the computational capabilities of edge networks. The integration of Large Language Models (LLMs) empowers EI to evolve into the next stage: Edge General Intelligence (EGI), enabling more adaptive and versatile applications that require advanced understanding and reasoning capabilities. However, systematic exploration in this area remains insufficient. This survey delineates the distinctions between EGI and traditional EI, categorizing LLM-empowered EGI into three conceptual systems: centralized, hybrid, and decentralized. For each system, we detail the framework designs and review existing implementations. Furthermore, we evaluate the performance and throughput of various Small Language Models (SLMs) that are more suitable for development on edge devices. This survey provides researchers with a comprehensive vision of EGI, offering insights into its vast potential and establishing a foundation for future advancements in this rapidly evolving field.

The dramatic increase in consumption of ultra-processed food has been associated with numerous adverse health effects. Given the public health consequences linked to ultra-processed food consumption, it is highly relevant to build computational models to predict the processing of food products. We created a range of machine learning, deep learning, and NLP models to predict the extent of food processing by integrating the FNDDS dataset of food products and their nutrient profiles with their reported NOVA processing level. Starting with the full nutritional panel of 102 features, we further implemented coarse-graining of features to 65 and 13 nutrients by dropping flavonoids and then by considering the 13-nutrient panel of FDA, respectively. LGBM Classifier and Random Forest emerged as the best model for 102 and 65 nutrients, respectively, with an F1-score of 0.9411 and 0.9345 and MCC of 0.8691 and 0.8543. For the 13-nutrient panel, Gradient Boost achieved the best F1-score of 0.9284 and MCC of 0.8425. We also implemented NLP based models, which exhibited state-of-the-art performance.