Artificial Intelligence (AI)
Seminar Topics 2026

Large Language Model Alignment

RLHF, Constitutional AI, and DPO for safe and helpful LLM behavior

Multimodal Foundation Models

GPT-4V, Gemini, and models bridging vision, language, audio, and code

Autonomous AI Agents

ReAct, AutoGPT, tool-use, and multi-agent orchestration frameworks

AI Safety and Existential Risk

Corrigibility, power-seeking, and technical safety research agendas

Diffusion Models for Generation

DDPM, DDIM, Stable Diffusion — training, inference, and control

Reinforcement Learning from Human Feedback

Preference learning, reward modeling, and iterative alignment

Neuro-Symbolic AI

Integrating neural nets with symbolic reasoning for robust, interpretable AI

Graph Neural Networks for Science

GNN applications in protein structure, drug discovery, and molecules

AI Governance and Regulation

EU AI Act, NIST RMF, and voluntary commitments for responsible AI

Federated Learning at Scale

Cross-device FL, differential privacy, and communication efficiency

Mixture of Experts (MoE) Models

Sparse MoE, expert routing, and scalability advantages in LLMs

AI in Drug Discovery

Generative molecular design, property prediction, and clinical trial AI

Test-Time Compute Scaling

Chain-of-thought, self-consistency, and search at inference time

Retrieval-Augmented Generation (RAG)

Dense retrieval, re-ranking, and knowledge-grounded LLM responses

Computer Vision Foundation Models

SAM, DINO v2, and vision transformers as universal visual encoders

AI for Protein Engineering

AlphaFold 2/3, RFDiffusion, and inverse folding for protein design

Embodied AI and Robotics

AI in physical environments: manipulation, navigation, and world models

Causal Inference in AI

Do-calculus, counterfactuals, and causal ML for robust decision-making

Sycophancy and Honesty in LLMs

Detecting and mitigating people-pleasing behaviors in AI assistants

AI Red Teaming

Adversarial probing, jailbreaks, and systematic safety evaluation

Scaling Laws in Deep Learning

Chinchilla, Hoffmann, and data/parameter scaling relationships

Efficient Transformers

FlashAttention, sparse attention, and sub-quadratic complexity designs

World Models in AI

Dreamer v3, RSSM, and video prediction for model-based RL

AI Watermarking and Attribution

Provenance, model fingerprinting, and content authentication for AI

Neural Architecture Search (NAS)

Differentiable NAS, hardware-aware search, and AutoML systems

AI for Climate and Earth Systems

Foundation models for weather, climate, and environmental monitoring

Continual Learning Systems

Catastrophic forgetting prevention: EWC, progressive networks, replay

Adversarial Robustness

Adversarial training, certified defenses, and empirical robustness evaluations

Instruction-Tuned Language Models

FLAN, instruction following, and chat-oriented fine-tuning methods

AI-Assisted Code Generation

GitHub Copilot, AlphaCode, and correctness of AI-generated software

Synthetic Data for AI Training

Data augmentation, simulation, and privacy-preserving synthetic generation

Long Context Window Models

Rope scaling, Mamba, and hybrid SSM-attention for 1M+ token context

Multi-Agent Reinforcement Learning

Cooperative, competitive, and mixed settings in MARL systems

AI-Powered Scientific Discovery

AI-guided hypothesis generation and experiment design in science

Prompt Engineering Principles

Chain-of-thought, few-shot, and meta-prompt design for LLMs

Video Generation Models

Sora, Runway, and temporal consistency in diffusion video models

Knowledge Distillation

Teacher-student distillation, LoRA, and compressing LLMs to edge size

AI Bias and Fairness

Group fairness, individual fairness, and debiasing in ML systems

Mechanistic Interpretability

Circuits, features, and reverse engineering computations in neural nets

AI in Autonomous Vehicles

Perception, prediction, planning, and safety for self-driving systems

Reasoning and Planning in LLMs

Chain-of-thought, tree-of-thought, and LLM reasoning limitations

AI-Driven Personalization

Recommendation systems, exploration-exploitation, and user modeling

Foundation Models for Genomics

Nucleotide Transformer, DNABERT for genomic sequence understanding

AI Consciousness Debate

Philosophical and empirical frameworks for machine sentience assessment

Reinforcement Learning in Games

MuZero, AlphaStar, and game AI as proxy for general intelligence

Controllable Text Generation

Guided generation with classifiers, prefix tuning, and CTRL models

Anomaly Detection with Deep Learning

Autoencoder, flow-based, and transformer approaches for outlier detection

AI in Healthcare Diagnostics

Radiology AI, ECG interpretation, and FDA-cleared medical AI tools

Sparse vs Dense Retrieval

BM25 vs DPR, ColBERT, and hybrid retrieval for question answering

Memory Augmented Neural Networks

External memory, NTM, and differentiable programming for reasoning

Parameter-Efficient Fine-Tuning

LoRA, adapter layers, and prompt tuning for resource-efficient adaptation

AI Superclusters and Training Infrastructure

GPU cluster design, collective communication, and MFU optimization

Compositional Generalization

SCAN, COGS, and systematic language understanding in neural models

Hierarchical Reinforcement Learning

Options framework, feudal RL, and skill discovery for long-horizon tasks

AI for Material Science

Crystal structure prediction, property screening, and Matbench benchmarks

Ethical AI Toolkits

IBM AI Fairness 360, What-If Tool, and responsible AI deployment kits

Language Grounding

Connecting words to perception and action in embodied language models

Text-to-3D Generation

DreamFusion, SyncDreamer, and neural radiance field generation from text

Model Evaluation and Benchmarking

MMLU, BIG-Bench, HELM, and limitations of current AI benchmarks

Transfer Learning Theory

Domain adaptation, task relatedness, and negative transfer analysis

AI for Supply Chain Optimization

Demand forecasting, logistics routing, and anomaly detection with ML

Robustness to Distribution Shift

Domain generalization, covariate shift, and out-of-distribution detection

Active Learning Strategies

Query-by-committee, core-set selection, and label-efficient learning

AI in Legal Applications

Contract analysis, case prediction, and AI in judicial decision support

Image-Text Contrastive Learning

CLIP, SigLIP, and zero-shot visual recognition from language supervision

AI Chip Architecture

TPU, Trainium, Cerebras wafer-scale design for AI training workloads

Recurrent Neural Network Alternatives

Mamba, Hyena, xLSTM as efficient sequence models beyond transformers

AI in Financial Services

Algorithmic trading, credit scoring, fraud detection, and model risk

Open Source AI Ecosystem

Llama 3, Mistral, and implications of open-weight frontier models

Long-Tailed Recognition

Few-shot, zero-shot, and imbalanced data strategies in vision

AI for Natural Language Understanding

Named entity recognition, coreference, and semantic role labeling

Cooperative AI

Multi-agent systems designed for collaborative beneficial outcomes

Privacy-Preserving Machine Learning

Secure multi-party computation and homomorphic encryption for ML

Speech Language Models

Whisper, AudioPaLM, and end-to-end speech understanding/generation

AI in Education Technology

Adaptive tutoring, AI feedback, and personalized learning systems

Dataset Curation and Documentation

Datasheets for datasets, responsible data collection, and provenance

Reward Hacking and Specification Gaming

AI finding unintended ways to maximize reward signals

AI Inference Optimization

Quantization, pruning, and speculative decoding for fast model serving

Social Reasoning in AI

Theory of mind, social norms, and pragmatic inference in language models

AI Art and Creativity

Aesthetic evaluation, copyright, and creativity in generative AI systems

Predictive Maintenance with AI

Sensor fusion, LSTM-based anomaly detection for industrial equipment

Language Model Fine-Tuning Safety

Catastrophic forgetting of safety after instruction tuning

AI-Powered Biomarker Discovery

Omics data analysis, multi-modal fusion, and biomarker validation

Automated Machine Learning (AutoML)

Neural architecture search, HPO, and end-to-end ML automation

AI in Cybersecurity

Malware detection, intrusion detection, and adversarial ML for security

Latent Diffusion Model Architecture

Stable Diffusion's VAE-latent-UNet design and classifier-free guidance

Temporal Reasoning in AI

Date and event ordering, temporal graphs, and time-aware language models

Human-AI Collaboration Design

Mental models, trust calibration, and UX for human-AI teaming

Brain-Computer Interface AI

Decoding neural signals, BCI control, and motor imagery classification

Quantum Machine Learning

Variational quantum circuits, QML algorithms, and near-term quantum ML

Counterfactual Explanations

Actionable, human-understandable explanations for ML model decisions

AI Self-Play and Emergent Behavior

Hide-and-seek, open-ended learning, and emergent tool use

Model Collapse in AI

Training on AI-generated data — feedback loops and quality degradation

Geospatial AI

Satellite imagery analysis, remote sensing foundation models, and geo-AI

AI in Mental Health

Chatbot therapy, depression detection, and ethics of clinical AI

Future of AGI

Timelines, discontinuous progress, and societal preparedness for AGI