Jayanth Ramakrishnan and Parvendan Rangaswamy, Department of Computer Science and Engineering, SRM University-AP, Amaravati, Andhra Pradesh, India
Large Language Models (LLMs) have gotten really good at generating code, and we usually measure thatwith Pass@k. But Pass@k only tells you how often the model gets it right on its best try. It says nothingabout how much the outputs vary or whether they’re structurally consistent across multiple runs. That’s abig problem for real-world use, where you typically call the model just once.So we came up with the Stochastic Stability Index (SSI). It’s a combined metric that looks at three things:whether the code is syntactically valid (Msyn), how similar its structure is to a correct solution using thenormalised Zhang–Shasha Tree Edit Distance on Abstract Syntax Trees (Mast), and whether it actuallyworks when run in a safe sandbox (Mfun). SSI directly penalises variation from one sample to the next,giving you a single reliability score that reflects actual deployment risk, not just best-case performance.We tested SSI on HumanEval and MBPP using three top code generation models—Qwen, DeepSeek-Coder-v2, and Yi-Coder-9b—at three temperatures (T ∈{0.2, 0.5, 0.8}), generating 20 independent samples perproblem (that’s n = 96,840 programs total). What we found is a Creativity–Stability Paradox that shows upacross all models and both benchmarks: Pass@10 goes up as you raise the temperature, but SSI goes downas outputs become more variable. For Qwen on HumanEval, Pass@10 jumps from 86.8% to 94.0%, whileSSI falls from 0.6833 to 0.6563. A one-way ANOVA on structural similarity across every model–benchmarkcombination (in all cases p > 0.05) shows that the drop comes almost entirely from functional inconsistency,not structural divergence. That tells us something important about how to design better stability metrics. Thebottom line: if you really care about production-ready code generation, you need reliability-aware metricsalongside Pass@k.
Large Language Models,Code Generation,Stochastic Stability Index,Abstract Syntax Trees,Pass@k,Inference Temperature,Benchmark Evaluation,Output Variance
Evariste Ntaryamira1, Grace Murhula Kabi2, 1Université Paris 8, Saint-Denis, France, 2Institut Supérieur de Commerce (ISC), Bukavu, République Démocratique du Congo
Educational Management Information Systems in Sub-Saharan Africa collect data from thousands of schools each year to feed the indicators required by Sustainable Development Goal 4 and the Continental Education Strategy for Africa. A recurring problem, however, undermines this process: as data travels up the administrative pyramid, each echelon that handles a file reformats it, reinterprets column names, and silently alters values. By the time a record reaches the national level, it often says more about the chain of people who handled it than about the school that produced it. This paper introduces a multi-layer semantic mediation architecture that addresses the problem at its root. Positioned between school-level data sources and the national server in a Hub-and-Spoke topology, the architecture comprises six formally defined processing layers: ingestion, statistical profiling, adaptive schema matching, coherence validation, incremental dictionary learning, and transactional synchronization. We formalize the mediation process as a structure-preserving functor between schema categories and prove its correctness under composition. Three original algorithms are proposed and analyzed: AdaptiveSchemaMatch, which combines TF-IDF lexical analysis, distributional divergence, and value-set overlap into a weighted similarity measure with gradientbased weight adaptation; IncrementalDictLearn, a reinforcement-inspired dictionary learner with provable convergence; and CoherenceGuard, a multi-tier consistency engine. We further show how cross-lingual sentence embeddings and lightweight autoencoders can augment the matching and anomaly detection capabilities in data-rich settings, while remaining deployable on resourceconstrained devices. The architecture is designed for offline-first, low-connectivity environments and scales to national-level volumes through a formally analyzed batch-parallel processing model. We describe the deployment context within the PAQABU project (UNESCO, Burundi) and the eSchool platform (Democratic Republic of Congo), and discuss the implications for educational data governance across the continent.
Semantic mediation, schema matching, educational data, Hub-and-Spoke, EMIS, incremental learning, data quality, Big Data, deep learning, Sub-Saharan Africa
Kazim Ali Korejo, Ana Maria Martinez-Enriquez, and Faaiza Zulfqar, Computer Science, Cinvestav-IPN, Av Instituto Politecnico Nacional 2508 Zacatenco, 07360, Mexico City, Mexico
Generative adversarial networks have emerged as a significant paradigm in Deep learning for Natural Language Processing, enabling improvements in multilingual learning, semantic modeling, text generation, and multimodal synthesis, to name a few. Our research aims to analyze the underlying learning mechanisms and the architectures used for different applications, such as paraphrase generation, text-toimage synthesis, machine translation, data augmentation, sentiment analysis, and noise reduction. Besides, our study analyzes the integration of Transformer with multilingual adaptation, focusing on limitations such as training instability, collapse mode, unreliable evaluation, and semantic inconsistency. Lastly, we outline opportunities for the way forward involving retrieval-augmented, explainable multimodal semantic alignment, symbolic reasoning, and hybrid large language model frameworks for more reliable, semantically grounded language generation.
GANs, Large Language Models, Multi-Agent, Generative Adversarial Networks, GAN-Variant.
Zihan Xu1 and Jason Moya2 1BASIS International School Hangzhou, No. 9 Yulin Road, Shangcheng District, Hangzhou, 310016, China 2California State Polytechnic University, Pomona, 3801 W Temple Ave, Pomona, CA 91768
Young athletes face elevated rates of anxiety, stress, and performance pressure, yet rarely access traditional sportpsychology services because of stigma, cost, and time scarcity during peak training periods. Athlete Emotion Manager is a Flutter and Firebase mobile application that consolidates four evidence-supported techniques — box breathing, guided visualization, grounding, and curated educational articles — into a single self-regulation platform built around three components: a Firebase-backed authentication system, a longitudinal mood-tracking and visualization system that surfaces a personalized trend chart, and a progress-and-streak engagement system designed to reinforce consistent practice without manipulative patterns. A ten-day in-the-wild trial with 24 competitive athletes showed that low-friction exercises such as box breathing dominated actual usage, and that more than 75% of participants rated every delivered technique as somewhat or very helpful. The application demonstrates that a focused, mood-aware mobile tool can meaningfully extend the reach of athlete mental-wellness support.
Athlete mental health, Sports psychology, Emotional self-regulation, Mobile health (mHealth), Firebase.
Vivienne Zhai1and Quincy Stokes2 1St. Margarets Episcopal School, 31641 La Novia Ave, San Juan Capistrano, CA 92675 2University of California, Irvine, Irvine, CA 92697
Many athletes struggle with recovery, rehabilitation, and long term wellness after injuries or physical strain. Athlete Well was created to provide athletes with recovery guidance, wellness education, and progress tracking in one organized platform. The app includes three major systems: recovery plans, progress tracking, and educational wellness resources. These systems help users follow rehabilitation exercises, monitor improvement over time, and learn about injury prevention, mobility, nutrition, rest, and mental wellness [1]. One challenge during development was organizing different recovery resources into a simple and easy-to-use interface. To evaluate the app, I conducted a survey experiment where users tested Athlete Well and rated its usability and helpfulness. The results showed that most users found the app organized, understandable, and useful for recovery support [2]. Athlete Well is important because it encourages athletes to recover safely, understand their bodies better, and build healthier long-term recovery habits instead of focusing only on athletic performance.
Sports Recovery, Rehabilitation Tracking, Athlete Wellness, Mobile Health Application
Junde Zhu1, Rodrigo Onate2, 1Army and navy academy, 2605 Carlsbaad Blvd Caarlsbad, CA 92008, California State University, Fullerton, 800 N State College Blvd, Fullerton, CA 92831
Food waste remains a significant problem because many people are unsure how to use ingredients already available in their kitchens [1]. This often results in unnecessary grocery purchases, wasted food, and unhealthy dining habits. To address this issue, this project developed YummySnap, a mobile application that uses artificial intelligence to identify ingredients from photographs and generate personalized recipes with nutritional information. The system was built using Flutter, OpenAI image analysis and language models, and local storage through shared_preferences [2]. Three major components were implemented: ingredient recognition, recipe generation, and recipe history management. Several challenges were considered, including image quality, user interface design, and data persistence. Experimental testing showed that image clarity strongly affects ingredient detection accuracy, while user preferences significantly improve recipe relevance. These findings demonstrate that combining computer vision with personalized AI recommendations can create a practical tool for reducing food waste and encouraging healthier home cooking.
Artificial Intelligence, Health, Simple, Waste management
Haoyu He1, Jonathan Sahagun2, 1Sage Hill School, 20402 Newport Coast Dr, Newport Coast, CA 92657, 2California State University, Los Angeles, 5151 State University Dr, Los Angeles, CA 90032
Compulsive “doomscrolling” harms attention and wellbeing, yet infinite feeds remove the natural cues that would prompt a user to stop. Focusrail is a wearable inertial sensor, worn on the hand, that detects the motion of scrolling and alerts a connected phone in real time. An LSM6DSOX IMU streams accelerometer and gyroscope data at 50 Hz over Bluetooth Low Energy to a Flutter application, which runs a 12-sample sliding-window detector and fires a gentle alert with a cooldown to avoid fatigue, logging each event for weekly review. Key challenges — separating scrolling from ordinary motion, reliable low-power streaming, and alert fatigue — were addressed through dual acceleration/gyroscope thresholds, BLE with auto-reconnect, and a suppression interval. Experiments showed scrolling is cleanly separable from resting, typing, and walking, and that the chosen gyroscope threshold sits at the optimum of the precision–recall curve. By cueing the behavior the instant it begins, Focusrail enables more intentional phone use.
Doomscrolling, Wearable sensing, Inertial measurement unit (IMU), Bluetooth Low Energy (BLE), Flutter
Jiaxuan Wang1, Bobby Nguyen2, 1Mercersburg Academy, 100 Academy Drive, Mercersburg, PA 17236 2University of California, Riverside, 900 University Ave, Riverside, CA 92521
Alzheimer’s disease affects millions of patients and burdens nearly 13 million unpaid caregivers in the United States alone, yet nonpharmacological support tools such as reminiscence therapy remain fragmented across paper memory books, standalone puzzle apps, and separate scheduling tools. MemoryWell proposes a single AI-supported mobile platform that turns family photos into interactive reminiscence puzzles, records positive “Shining Moments,” tracks caregiver mood and wellness, and coordinates care through a shared, real-time Care Circle. The system links a Flutter client, a Firebase real-time backend, and an OpenAI GPT-4o-mini-AI layer that generates personalized encouragement, pattern insights, and wellness tips. Implementation challenges included calibrating puzzle accessibility, constraining AI-generated content, and securing a shared real-time backend. Two experiments showed that puzzle difficulty grows super-linearly with grid size, validating the shipped 3×3 default, and that AI content reliability degrades directly with API failure rate due to an absence of retry logic. MemoryWell offers a more integrated, personalized alternative to prior single-purpose reminiscence, wellness, and coordination tools.
Dementia Care, Reminiscence Therapy, Mobile Health (mHealth), Artificial Intelligence, Large Language Models
Copyright © SIPR 2026