We present Hexaïssa, a novel framework for adaptive chess engine routing that formulates expert selection as a Mixture-of-Experts (MoE) problem. Hexaïssa learns a gating policy that dynamically selects among heterogeneous state-of-the-art engines—such as Stockfish, LCZero, and Obsidian—based on the tactical and strategic complexity of each board state. This adaptivity enables stronger play and more efficient computation than any fixed engine or static configuration. However, training such a gating policy is fundamentally challenging due to sparse optimization signals and long-horizon credit assignment in chess games. To address these challenges, we introduce a score-based inverse reinforcement learning (IRL) method that models expert engine trajectories as samples from a latent distribution over optimal behaviors. By recovering the Stein score function of this distribution via stochastic differential equations (SDEs), we infer dense, per-move reward signals consistent with potential-based IRL. These latent rewards allow efficient training of the gating network without requiring additional environment interaction or human supervision. Empirical results on standard chess benchmarks demonstrate that Hexaïssa significantly outperforms individual engines, conventional MoE models, and IRL baselines.

Multiple-choice questions (MCQs) serve as fundamental assessment tools in computing education, where high-quality distractors are critical for evaluating conceptual understanding and debugging skills.While large language models (LLMs) show promise in automating distractor generation, their effectiveness in reasoningintensive programming domains remains understudied.Another challenge is that current evaluation metrics often emphasize surfacelevel semantics rather than the logical reasoning required in programming tasks, limiting their practical utility for educators.To compare AI-generated and human-authored distractors, this study collected 925 MCQs from two online high school courses.The collected data include the question stem, correct answer, and three human-authored distractors for each question.For AI-generated distractor generation, we employed the GPT-4 API through a structured prompt containing: (1) question stem, (2) correct answer, (3) Bloom's taxonomy level, and (4) instructional constraints.To determine the Bloom's taxonomy level for each question, two assessment experts independently classified all questions based on Bloom's taxonomy (Remember, Understand, Apply, Analyze, Evaluate, Create), achieving moderate inter-rater reliability (Cohen's 𝐾 = 0.67).Discrepancies, which occurred in 33% of cases, were resolved by a third expert to ensure accurate cognitive-level categorization.The GPT-4 model generated three plausible distractors per question while maintaining cognitive-level alignment.This study proposes a human-AI collaborative framework to evaluate distractor quality in programming education.Our human-AI collaborative evaluation framework combined human expertise with AI analysis (using GPT-4 and DeepSeek-V3) through a three-phase process: First, human-created and AI-generated distractors were anonymized and randomized.Next, human experts and AI models independently selected the three most pedagogically effective distractors per question based on plausibility and challenge potential.Finally, we implemented a ranking system prioritizing distractors with the highest selection frequency across evaluators, with ties resolved by cross-validator agreement.Results demonstrate that AI-generated distractors achieve comparable quality to human-crafted ones for foundational programming concepts (e.g., syntax recall and basic logic).However, significant gaps emerge in higher-order cognitive domains, particularly

We study the Quality of Service Degradation (QoSD) problem, in which an adversary perturbs edge weights to degrade network performance. This setting arises in both network infrastructures and distributed ML systems, where communication quality, not just connectivity, determines functionality. While classical methods rely on combinatorial optimization, and recent ML approaches address only restricted linear variants with small-size networks, no prior model directly tackles the QoSD problem under nonlinear edge-weight functions. This work proposes Hephaestus, a self-reinforcing generative framework that synthesizes feasible solutions in latent space, to fill this gap. Our method includes three phases: (1) Forge: a Predictive Path-Stressing (PPS) algorithm that uses graph learning and approximation to produce feasible solutions with performance guarantee, (2) Morph: a new theoretically grounded training paradigm for Mixture of Conditional VAEs guided by an energy-based model to capture solution feature distributions, and (3) Refine: a reinforcement learning agent that explores this space to generate progressively near-optimal solutions using our designed differentiable reward function. Experiments on both synthetic and real-world networks show that our approach consistently outperforms classical and ML baselines, particularly in scenarios with nonlinear cost functions where traditional methods fail to generalize.