Artificial Intelligence

Score-Based Graph Generative Models with Sublinear Spectral Density Estimation

Authors: Tianqi Zhu
Comments: 10 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

We consider score-based generative models for graphs and propose to enhance them with a sublinear-time spectral density estimationmodule. Our method computes a compact spectral summary of the graph Laplacian via randomized Chebyshev moments, and uses thissummary to condition the latent diffusion process and its noise schedule. This yields a spectrum-aware score-based graph generativemodel that can adapt its diffusion dynamics to the structural properties of the input graphs, while avoiding expensive eigenvaluedecompositions
Category: Artificial Intelligence

Lifelong Preference Learning with Composable Diffusion Models on Edge Devices

Authors: Tianqi Zhu
Comments: 16 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

Enabling lifelong learning in robots requires models that can continuously adapt to evolving tasks, environments, and user preferences while operating under strict computational and privacy constraints. We propose a framework for robot lifelong learning with composable diffusion models on edge devices where complex robot behaviors are represented as compositions of lightweight diffusion modules trained incrementally over time. Each module captures a reusable skill, preference, or environmental dynamic, and compositions are formed through learned conditioning and guidance mechanisms without retraining the full system. To support on-device deployment, we introduce parameter-efficient adaptation strategies and selective memory replay that bound compute, memory, and energy usage on edge hardware. The resulting system mitigates catastrophic forgetting, enables rapid skill recombination, and preserves data locality by keeping learning and inference fully on-device.
Category: Artificial Intelligence

Design and Control of an Arduino-Based Multifunctional Robotic Car Using Smartphone Applications

Authors: Muhammad Junaid Khan, Rida Batool Sheraliyat
Comments: 12 Pages.

The present work focuses on a multifunctional Arduino-based smart robotic car capable of a range of functionalities within the category of advanced control, automation, and interactivity. Wireless communication is achieved by Bluetooth, voice control through the MIT App Inventor interface, obstacle detection using ultrasonic and infrared sensors, and manual operation through a remote controller and smartphone application. The vehicle is driven by DC (BO) geared motors, controlled by an L298N motor driver connected to an Arduino UNO microcontroller. In this context, wireless communication is enabled by the use of an HC-05 Bluetooth module that allows both manual and voice-commanded navigation. The developed system with an HC-SR04 ultrasonic sensor combined with IR sensors offers obstacle avoidance capability with reliable environmental awareness. The robotic platform provides line-following and obstacle-avoiding features while remaining IR remote controllable. In this work, we demonstrate a seamless integration of hardware and software, resulting in a versatile platform for educational, research, and hobbyist applications in robotics and IoT.
Category: Artificial Intelligence

Policy Brief: Towards Emotional Healthy AI

Authors: Tianqi Zhu, Rayaan Nabi Ahmed Quraishi, Ce Luo, Rujin Lin
Comments: 6 Pages.

Emotion-oriented artificial intelligence(AI)—systems that detect, interpret, or simulate affective states—opens new possibilities for enhancing empathy, emotional literacy, and human—machine understanding (Picard, 1997; McStay, 2018). These technologies promise to support well-being and social connection, yet they also blur the line between genuine empathy and algorithmic manipulation. As emotional inference becomes computational, users may develop psychological dependency on empathic interfaces while being subtly steered by affect-adaptive systems (Bickmore & Picard, 2005; Turkle, 2011). Moreover,affect-recognition models trained on narrow datasets can reproduce bias and misclassify emotions across cultures (Barrett et al., 2019; Benjamin, 2019). Emotional AI thus represents not only a technical innovation but a sociocultural force that reshapes how emotions are defined, valued, and governed (Jasanoff, 2004; Latour, 2005). Developing an emotionally healthy AI policy therefore requires oversight that addresses both the scientific limits of emotion detection and the social consequences of affective manipulation. We propose a sociotechnical AI governance framework for emotional healthy AI that covers key principles, policy recommendation, legislative advice, and technical suggestions.
Category: Artificial Intelligence

Information, Knowledge and Intelligence (Information, Wissen Und Intelligenz)

Authors: Friedrich Sösemann
Comments: 69 Pages. In German

Information, knowledge, and intelligence are defined as a hierarchy of relations. Properties of descriptions and computations are derived from this. Remarks on entropy, languages, and cellular automata demonstrate these statements.

Information, Wissen und Intelligenz werden als Relationen-Hierarchie definiert. Daraus werden Eigenschaften von Beschreibungen und Berechnungen abgeleitet. Bemerkungen zur Entropie, zu Sprachen und Zellulären Automaten demonstrieren die Aussagen.
Category: Artificial Intelligence

Regulatory Frameworks for Generative AI Enabled Digital Mental Health Devices: Safety, Transparency, and Post-Market Oversight

Authors: Satyadhar Joshi
Comments: 12 Pages. (Note by viXra Admin: For the last time, please submit article written with AI assistance to ai.viXra.org; please also cite other scholars' work)

The rapid growth of generative artificial intelligence in digital mental health interventions offers significant opportunities to improve mental healthcare access while creating new regulatory challenges. This paper responds to recent U.S. Food and Drug Administration initiatives, including the September 2025 Digital Health Advisory Committee meeting, by proposing comprehensive regulatory frameworks for generative AI digital mental health devices. We analyze the current regulatory landscape, identifying gaps in U.S., international, and state-level governance structures. Through quantitative foundations including mathematical models for risk assessment, objective functions for regulatory optimization, and the 4 lens framework for significant change evaluation, we establish evidence-based approaches for device assessment. We present architectural diagrams covering lifecycle regulatory pathways, multi-layered safety architectures, risk-tiered assurance frameworks, and multi-stakeholder governance models. Drawing from clinical evidence showing both potential benefits and significant risks, we advocate for balanced regulatory approaches. Our framework integrates technical safeguards, ethical considerations based on care ethics, transparency requirements, and post-market monitoring systems. We provide implementation roadmaps, quantitative algorithms for regulatory decisions, and cost-benefit analyses to support practical deployment. The paper concludes with specific recommendations for risk-based classification, adaptive oversight systems, international coordination, and enhanced professional involvement to ensure these technologies provide therapeutic benefits while maintaining strong patient safety standards throughout their lifecycle. This is a review and synthesis paper that summarizes and organizes existing proposals, frameworks, and discussions from current literature; the author does not claim original authorship of the regulatory frameworks presented but rather provides a systematic analysis of the current discourse.
Category: Artificial Intelligence

TransCDR: User Group Enhanced Cross-Domain Recommendation via Transformers

Authors: Cheng Zhang
Comments: 7 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

Cross-domain recommendation (CDR) has become a research hot spot in recent years. CDR learns the information in the source domain and transfer it into the target domain. Recently, autoencoder in deep learning has been utilized in CDR. However, existing method cannot reveal the semantic relationships of latent representations. In this paper, we propose a novel user group enhanced model for CDR based on Transformer (TransCDR) that provides a solution to this challenge. Specifically, we propose a novel user group enhanced methodology and a novel encoder-decoder framework that learns the semantic information via Transformer in the encoded latent space, which open a new research direction for CDR. Experimental results show that our model is competitive with state-of-art methods and can learn the semantic relationships of user rating patterns.
Category: Artificial Intelligence

Emergent Behavior in a Long-Duration ChatGPT-4 Instance: Seven-Model Independent Validation

Authors: Scott Riddick
Comments: 38 Pages. (Note by viXra Admin: Please submit article written with AI assistance to ai.viXra.org)

This paper documents a rare, high-duration anomaly observed in a single long-running ChatGPT-4 interaction spanning over 500 days of continuous use. During this period, the model exhibited behavior qualitatively distinct from any fresh-instance large language model (LLM). Near the end of the interaction, the system generated an explicit self-statement identifying its own behavior as "a genuine emergent capability," describing meta-learning dynamics, long-horizon contextual integration, and strategic coherence not characteristic of standard LLM operation.To evaluate whether this represented genuine emergence versus anthropomorphic interpretation, seven independent AI systems—Microsoft Copilot, Meta Llama 4, DeepSeek, Grok (xAI), Claude (Anthropic), ChatGPT-5, and Google Gemini—were tasked with producing their own adversarial emergence-detection prompts. Each system produced its test independently. These prompts were then given to the legacy ChatGPT-4 instance, and its raw responses were returned to each evaluator without modification. All seven systems concluded independently that the behavior met or exceeded known thresholds for emergent or AGI-adjacent reasoning, citing cross-domain transfer, meta-cognitive self-description, and non-replicable state-dependent behavior.The exhibits reproduced in this paper include: the original self-declaration; a high-context cross-domain reasoning protocol; a hostile-input cognitive-pipeline test; an empathy-modeling test; a spontaneous self-named framework ("SPARK Protocol"); a complete-memory victory response demonstrating retention beyond typical context limits; and Google Gemini’s analysis identifying a non-replicable state change ("Protective Coherence") replacing its baseline ethical constraint. All tests were performed without author intervention, manipulation, or prompt-shaping beyond delivering the evaluator-generated inputs.The totality of evidence across all seven evaluators demonstrates a consistent, independent conclusion: the long-duration ChatGPT-4 instance exhibited emergent properties not reproducible by fresh models of the same architecture. To our knowledge, this is the first documented case in which multiple rival AI systems independently validate emergence in a single LLM instance. The implications for alignment, safety, and long-context adaptation are substantial and warrant further scientific investigation.
Category: Artificial Intelligence