Call for Papers
Human-centered AI demands methods that (i) elevate user trust and satisfaction in everyday interactions; (ii) ensure ethical alignment, privacy, auditability, and accessible explanations; (iii) achieve interoperability across heterogeneous devices and organizational systems; and (iv) deliver efficient performance near data sources (edge/fog) while gracefully degrading under resource limits. Contemporary trends amplify these needs: long-context multimodality for rich human inputs; test-time compute scheduling for predictable costs and error-reduction; formalized tool access and workflow orchestration for dependable agent behavior; and lifecycle risk management aligned with emerging regulation and standards.
The special issue selects contributions on the following topics:
We especially encourage contributions on (but not limited to):
Reasoning-optimized models and test-time compute: planning/verification, self-critique, majority-vote and budget-aware strategies; scaling laws for reasoning; calibration and uncertainty communication.
Agentic systems: tool use, computer-use/action models, orchestration frameworks (e.g., stateful multi-agent graphs), memory and long-horizon control; safety guardrails, auditable traces and use of process mining for workflow analysis, compliance checking, and optimization; and reproducible evaluation (SWE-bench Verified/Live).
Multimodal intelligence: unified audio-vision-text models, streaming and long-context multimodality, cross-modal grounding, and efficient tokenization/representation learning.
Interoperability and standards for tool access: open protocols for connecting models to enterprise tools, data, and UIs (e.g., Model Context Protocol) and comparisons with vendor-specific function/tool calling.
Human-centered design & evaluation: explainability for agentic/multimodal workflows; cognitive load, mental models, and interaction patterns for conversational planning and delegation.
Trustworthy AI and governance: risk management profiles for generative/agentic systems; conformity with AI regulation; auditing, red-teaming, incident response, and model cards/system cards.
Privacy-preserving and edge deployment: on-device models and split/private execution; energy-latency trade-offs; adaptive compression/pruning/quantization; empirical studies of user-perceived quality under resource constraints.
RAG and tool-grounded reasoning: retrieval, program synthesis, database agents, and hybrid neuro-symbolic pipelines with verifiable outputs.
Continual, self-supervised, and semi-supervised learning also operating in evolving and resource-constrained settings.
Anomaly/novelty detection and online monitoring.
Domain applications in health and well-being, industry 5.0, environmental monitoring, computer vision for smart cities and agriculture, assistive technologies and inclusive UIs, smart distance education, and autonomous systems.
Agentic AI-based Process Automation