Research

CyberWorld: World Models for Sample-Efficient Autonomous Cyber Defense

Ryozo Masukawa, Sanggeon Yun, Raheeb Hassan, Hyunwoo Oh, Sungheon Jeong, Mohsen Imani

A Dreamer-style world model for autonomous cyber defense that reaches a strong control policy in thousands of environment steps instead of millions.

Vector Symbolic Policy Gradient

Ryozo Masukawa, Sanggeon Yun, Sungheon Jeong, Hyunwoo Oh, Raheeb Hassan, Pietro Mercati, Nathaniel D. Bastian, Mahdi Imani, Mohsen Imani

An actor that represents each action as a hypervector; its policy-gradient update is exactly advantage-weighted bundling, with provable robustness to bit flips.

Qubit-Efficient Quantum Search for Hyperdimensional Decomposition via Logarithmic Encoding

Sanggeon Yun, Hyunwoo Oh, Ryozo Masukawa, Raheeb Hassan, Mohsen Imani

Encodes hypervectors in O(log D) qubits for quantum hypervector decomposition, using up to 2,000x fewer qubits while keeping the quadratic search speedup.

n-Musketeers: Reinforcement Learning Shapes Collaboration Among Language Models

Ryozo Masukawa, Sanggeon Yun, Hyunwoo Oh, Sungheon Jeong, Raheeb Hassan, Hanning Chen, Wen-Jun Huang, Mahdi Imani, Pietro Mercati, Nathaniel D. Bastian, Mohsen Imani

Frozen small language models collaborate through a trainable attention interface over their hidden states, trained with RL from verifiable rewards.

HopFormer: Sparse Graph Transformers with Explicit Receptive Field Control

Sanggeon Yun, Raheeb Hassan, Ryozo Masukawa, Sungheon Jeong, Mohsen Imani

A graph Transformer that injects structure only through head-specific n-hop sparse attention, with no positional encodings and cost linear in mask sparsity.

Contextual Fusion Strategies for Multimodal GNN-Based Reasoning: Performance and Computational Trade-Offs

Sanggeon Yun, Ryozo Masukawa, Raheeb Hassan, Minhyoung Na, Mohsen Imani

Three knowledge-graph-centric fusion strategies for multimodal video anomaly detection, reaching 72.93% mAP on XD-Violence; early fusion cuts per-frame energy by over 70%.

MissionHD: Hyperdimensional Refinement of Distribution-Deficient Reasoning Graphs for Video Anomaly Detection

Sanggeon Yun, Raheeb Hassan, Ryozo Masukawa, Nathaniel D. Bastian, Mohsen Imani

Refines LLM-generated reasoning graphs directly in hyperdimensional space, improving weakly supervised video anomaly detection and recognition.

Reinforcement Learning Based Formulations with Hamiltonian-Inspired Loss Functions for Combinatorial Optimization over Graphs

Redwan Ahmed Rizvee, Raheeb Hassan, Md. Mosaddek Khan

Uses the QUBO Hamiltonian as an RL reward for combinatorial optimization over graphs, improving on PI-GNN by up to 44%.

DePAint: A Decentralized Safe Multi-Agent Reinforcement Learning Algorithm considering Peak and Average Constraints

My first first-author paper.

Raheeb Hassan, K.M. Shadman Wadith, Md. Mamun or Rashid, Md. Mosaddek Khan

A privacy-preserving, fully decentralized multi-agent policy gradient method that satisfies both peak and average safety constraints.