Student Research Topics

Explore research directions at SigM Lab, Hanoi University of Science and Technology. Find an area that matches your interests and contact the listed supervisors to discuss a suitable topic.

Nonlinear Signal Processing

Assoc. Prof. Manh-Thang Hoang

Complex networks for pattern recognition

From EEG signals to network features and classification.
From EEG signals to network features and classification.

Study complex networks and their applications to physiological signals, including PPGs and EEGs.

Areas: Complex networks, signal analysis

Contact: thang.hoangmanh@hust.edu.vn

Detecting regime changes in nonlinear signals

Signal changes revealed through nonlinear time-series analysis.
Signal changes revealed through nonlinear time-series analysis.

Explore nonlinear analysis methods such as MF-DFA, EMD and EKF/UKF to detect changes or switching in signal behavior.

Areas: Time series, nonlinear analysis

Contact: thang.hoangmanh@hust.edu.vn

Chaotic cryptography on FPGA

An FPGA-based image encryption and transfer workflow.
An FPGA-based image encryption and transfer workflow.

Design chaotic cryptographic systems and investigate cryptanalysis with FPGA implementation.

Areas: Cryptography, nonlinear systems, FPGA

Contact: thang.hoangmanh@hust.edu.vn

Computer Vision and Multimodal Learning

Assoc. Prof. Thi-Lan Le · Assoc. Prof. Thanh-Hai Tran · Assoc. Prof. Hai Vu

Human action recognition

Recognizing actions for exercise assessment and human–robot interaction.
Recognizing actions for exercise assessment and human–robot interaction.

Action classification and detection, including continuous action recognition. Applications include human–COBOT interaction and exercise assessment.

Areas: Video understanding, human–robot interaction

Contact: lan.lethi1@hust.edu.vn, hai.tranthithanh1@hust.edu.vn, hai.vu@hust.edu.vn

Medical data and image analysis

An example of medical image analysis using anatomical landmarks.
An example of medical image analysis using anatomical landmarks.

Detect and segment lesions or tumors in endoscopy and ultrasound images, and classify anatomical landmarks to support doctors.

Areas: Medical imaging, detection, segmentation

Contact: lan.lethi1@hust.edu.vn, hai.tranthithanh1@hust.edu.vn, hai.vu@hust.edu.vn

Text-based person search

Matching a natural-language description with person images.
Matching a natural-language description with person images.

Align natural-language descriptions with person images using text and image encoders and vision-language models.

Areas: Vision-language models, image retrieval

Contact: lan.lethi1@hust.edu.vn, hai.tranthithanh1@hust.edu.vn

Robust federated learning

Federated learning with distributed clients and noisy labels.
Federated learning with distributed clients and noisy labels.

Study client selection, label-noise handling and communication efficiency, with healthcare applications.

Areas: Federated learning, learning with noisy labels

Contact: lan.lethi1@hust.edu.vn, hai.tranthithanh1@hust.edu.vn

Speech Processing and Multimodal LLMs

Dr. Ngoc-Diep Do

Speech processing

Speaker verification and countermeasures against spoofed speech.
Speaker verification and countermeasures against spoofed speech.

Investigate speaker verification and speech spoofing detection.

Areas: Speech, audio analysis, authentication

Contact: diep.dothingoc@hust.edu.vn

Multimodal LLMs

A multimodal language model connects audio, text, images and video.
A multimodal language model connects audio, text, images and video.

Explore large language models that combine text, speech and image tokenizers.

Areas: Language models, speech, multimodal learning

Contact: diep.dothingoc@hust.edu.vn

Secure and Trustworthy Medical AI

Dr. Kim-Hue Ta

Access control for Fog-based IoMT

Medical record sharing with attribute-based access control and verification.
Medical record sharing with attribute-based access control and verification.

Explore multi-authority ciphertext-policy attribute-based encryption (CP-ABE) and zero-knowledge proofs of attribute ownership for sharing medical records across hospitals.

Areas: Access control, CP-ABE, zero-knowledge proofs

Contact: hue.tathikim@hust.edu.vn

DICOM image integrity

Verifying DICOM image integrity with embedded evidence.
Verifying DICOM image integrity with embedded evidence.

Study reversible embedding in non-diagnostic image borders with chaotic scrambling and Groth16 zk-SNARK verification to detect tampering.

Areas: Medical data integrity, steganography, zk-SNARKs

Contact: hue.tathikim@hust.edu.vn

Verifiable clinical AI inference (zkML)

Private, verifiable inference and clinical evidence binding.
Private, verifiable inference and clinical evidence binding.

Prove that a declared model produced a clinical result while protecting patient data and model weights. Explore VOLE-based proofs and links to clinical evidence.

Areas: Zero-knowledge machine learning, clinical AI

Contact: hue.tathikim@hust.edu.vn

Post-quantum security for medical data and 5G

Combining post-quantum cryptography with verified physical-layer keys.
Combining post-quantum cryptography with verified physical-layer keys.

Explore lattice-based CP-ABE/ZKP and hybrid ML-KEM/ML-DSA with physical-layer keys admitted only after verification tests.

Areas: Post-quantum cryptography, physical-layer security

Contact: hue.tathikim@hust.edu.vn

Model lineage and LLM provenance

Studying model lineage, evasion and provenance.
Studying model lineage, evasion and provenance.

Study model lineage and detector evasion. Explore watermarking, provenance auditing and private LLM inference.

Areas: Trustworthy AI, provenance, model security

Contact: hue.tathikim@hust.edu.vn

These directions include ongoing work under review, work in preparation and exploratory studies. Model lineage and LLM provenance is an early-stage direction open to students.

LLMs, LVLMs and Edge AI

Dr. Tien-Cuong Bui

Trustworthy LLM and LVLM judges

Example A: gloves on the bench, with bare hands.Example B: blue gloves worn on both hands.
Compare whether the evidence supports a claim about wearing gloves.

Check model answers against text, images and video. Compare judging methods using accuracy, agreement with human labels and calibration.

Areas: LLM evaluation, visual question answering, grounded RAG

Contact: cuong.buitien@hust.edu.vn

Video event understanding with LVLMs

Tracking a person and bag over time to verify an event.
Tracking a person and bag over time to verify an event.

Combine detection, tracking and temporal reasoning to retrieve and verify video events, such as a bag left behind.

Areas: Video retrieval, tracking, temporal reasoning

Contact: cuong.buitien@hust.edu.vn

Multimodal agents for smart monitoring

An agent checks camera, sensor and log evidence before escalating.
An agent checks camera, sensor and log evidence before escalating.

Build agents that check camera, sensor and log evidence before sending alerts or escalating to a human reviewer.

Areas: Multimodal agents, monitoring, tool use

Contact: cuong.buitien@hust.edu.vn

Efficient LVLM/VLA inference on Jetson

Scheduling multiple vision workloads on a shared Jetson device.
Scheduling multiple vision workloads on a shared Jetson device.

Schedule vision and language models on Jetson devices. Explore adaptive visual tokens, multiple camera streams and real-time control.

Areas: Edge AI, embedded inference, scheduling

Contact: cuong.buitien@hust.edu.vn

Meet the team · Research overview · Lab projects