Adaptive Learning Systems

📄 Full text: arXiv:2604.26962 · local · arXiv:2409.03512 · local · arXiv:2507.12484 · local

Systems that dynamically adjust instructional content, difficulty, and feedback based on real-time learner models. The transition from static rule-based adaptation to agent-native, memory-driven personalization is a defining trend in 2024–2026.

From Static to Agent-Native Adaptation

Conventional tutoring systems rely on pre-trained knowledge with fixed difficulty ladders. RAG-augmented systems retrieve relevant content but lack granular learner profiles. Agent-native systems (e.g., DeepTutor) build a continuously evolving personalization substrate into every feature.

DeepTutor: Agent-Native Personalization

Ma et al. (2026) propose a closed tutoring loop that bidirectionally couples:

Citation-grounded problem solving — Ensures factual accuracy

Difficulty-calibrated question generation — Dynamically adjusts challenge level

A hybrid personalization engine combines static knowledge grounding with dynamic multi-resolution memory, distilling interaction history into a learner profile. The TutorBot layer deploys proactive multi-agent skills through extensible, cross-platform channels.

MAIC: Scale + Adaptivity

Yu et al. (2024) frame the core challenge as balancing massive scalability (MOOC-era reach) with personalized adaptivity (tutoring-era depth). Their LLM-driven multi-agent classroom at Tsinghua uses multiple archetype agents to create adaptive social contexts rather than only adapting content difficulty.

Chudziak & Kostka (2025): Multi-Agent Math Platform

Chudziak & Kostka (2025) introduce a multi-agent math tutoring platform combining three components:

1. Adaptive and personalized feedback — tailored to individual student needs 2. Structured course generation — curriculum-aligned learning paths 3. Textbook knowledge retrieval — grounding in established pedagogical sources

The platform enables modular, tool-assisted learning: students learn new topics while identifying weaknesses, revise for exams, and practice unlimited personalized exercises. Unlike reactive chatbots that answer ad-hoc questions, this system proactively structures the entire learning trajectory.

Research question: How can AI tutoring systems move beyond reactive assistance to enable structured, individualized, tool-assisted learning experiences? The answer involves combining agentic planning (course generation) with adaptive execution (personalized feedback and practice).

Design Principles

1. Learner-profile centrality — Every feature reads from and writes to the same profile 2. First-person evaluation — Measure adaptivity from the learner's perspective, not the system's 3. Cross-modality coherence — Adaptation should be consistent across text, code, and visual explanations 4. Proactive vs. reactive — Move from answering questions to anticipating learning needs

stanbkt-bayesian-knowledge-tracing — Posterior inference from BKT enables principled adaptive interventions

ai-enabled-serious-games — Extends adaptive learning principles to game-based training across professional domains

simulating-learner-task-selection — Quantifies how learner autonomy interacts with mastery learning algorithm optimization

expert-cognition-dashboard — Incorporates expert-like cognitive reasoning into adaptive system architectures

mbp-kt-meta-behavioral-knowledge-tracing — Model-agnostic injection strategies for KT architectures

understanding-student-effort-response-time — Understanding Student Effort Using Response-Time Propensitie...

llm-educational-simulation-adhd — Simulation-based testing for neurodivergent-adaptive system design

moodle-ai-tutoring-deep-learning — Emphasizes qualitative shifts in understanding over quantitative difficulty

ecnuclaw-k12-personalized-companion — ECNUClaw operationalizes real-time learner profiling in dialogue-based tutoring

cognitive-agent-compilation — CAC enables adaptive behavior grounded in explicit, editable learner models

ai-adult-learning-design — 19 design guidelines for AI technologies supporting adult learners

multimodal-learning-genai — Multimodal personalization across the six engagement types

ai-tutor-effectiveness-review — Adaptive algorithms in the mixed-effectiveness landscape

authentic-assessment — Beyond content adaptivity to assessment co-design

personalized-learning — The pedagogical goal that adaptive systems serve

knowledge-tracing-irt — Statistical models that inform adaptation decisions

agentic-workflows-education — Agentic planning paradigms for adaptation

human-in-the-loop-ai — Human validation of adaptive trajectories

llm-student-modeling-memory — Longitudinal memory architectures for learner profiles

collaborative-ai-tutoring — Adaptive systems for pair and group learning

affective-tutoring — Emotional state as input for adaptive difficulty calibration

socratic-ai-dialogue — Structured Socratic scaffolding as adaptive intervention

learnmate2-llm-adaptive-learning — Integrated system with PlanMate, StudyMate, AdaptMate components

k-12-ai-education — K-12 adaptive platforms

quantum-education-its — Knowledge-graph augmented ITS for quantum computing education

neural-symbolic-knowledge-tracing — Interpretable learner models for adaptivity

pedagogical-safety-rl — RL as personalization mechanism

engagement-assessment-video — Real-time engagement tracking for adaptivity

text-simplification-its — Content adaptation via text simplification

ai-metacognition-stem-review — ALEKS, Carnegie Learning as adaptive platforms for metacognitive outcomes

engagement-intensity-learner-modeling -- Engagement intensity during AI ethics instruction serves as an effective learner...

adapt-adaptive-lesson-plan-transformer -- AdaPT uses transformers to adapt lesson plans across regional and differentiated...

agentic-ai-education-scoping-review — Wang et al. (2026) scoping review: 474 studies on agentic AI in education, capability dimensions, and the frontier-agent technology gap

Sources

Ma et al. (2026). DeepTutor: Towards Agentic Personalized Tutoring. arXiv:2604.26962. PDF

Yu et al. (2024). From MOOC to MAIC: Reshaping Online Teaching and Learning through LLM-driven Agents. arXiv:2409.03512. PDF