| Nama Dosen | No. | Judul Topik | Deskripsi Singkat | Jenis |
|---|---|---|---|---|
| Erdefi Rakun & Kurniawati Azizah | 1 |
Bisindo Sign Language Recognition and Translation Using INews Dataset and state of the art models
|
[Gesture-to-Text Bisindo] Catatan: Dataset yang diturunkan dari siaran berita INews telah dikerjakan sebagai TA Kelompok pada sms 2025-2026 genap |
TA (S1) |
| 2 |
Transformer-Based Text Translation for SIBI Sign Language Animated Subtitles using TVRI Dataset
|
[Text-to-Gesture SIBI] Catatan: Proses pembangkitan animasi 3D berdasarkan kalimat SIBI sudah tersedia |
TA (S1) | |
| 3 |
Imagined Speech Event Related Detection
|
[Imagined Speech] Dalam penelitian ini, mahasiswa S1 diharapkan dapat mengembangkan metode untuk mendeteksi dan melakukan segmentasi temporal pada sinyal EEG mentah (raw EEG) guna mengidentifikasi transisi antar kondisi (rest → cue → imagined speech). Input berupa continuous/raw EEG signal yang masih mengandung berbagai fase eksperimen, sedangkan output yang diharapkan adalah label temporal atau timestamp segment yang menunjukkan kapan masing-masing kondisi terjadi. Pendekatan dapat melibatkan preprocessing (filtering, artifact removal), ekstraksi fitur berbasis waktu/frekuensi, serta metode classification, change-point detection, atau analisis event-related (misalnya ERD/ERS). Prasyarat yang diperlukan mencakup pemahaman pengolahan sinyal, machine learning (klasifikasi dan evaluasi model), serta dasar EEG/BCI. |
TA (S1) | |
| 4 |
Impact of EEG Preprocessing Strategies and Psychiatric Assessment on Imagined Speech Classification Performance
|
[Imagined Speech] Prasyarat yang diperlukan mencakup pemahaman pengolahan sinyal, machine learning (klasifikasi dan evaluasi model), serta dasar EEG/BCI |
Thesis (S2) | |
| Chan Basaruddin & Tusty Nadia Maghfira | 1 |
Multimodal social signal processing & applications
|
Topik ini mencakup berbagai teknik pemrosesan sinyal sosial multimodal untuk berbagai bidang terapan berbasis pembelajaran mesin, seperti pengenalan emosi, pengenalan stress atau gangguan mental, elderly care, infant care, crowd analysis, social attachment, dll. Tantangan utama terkait dengan teknik pemrosesan data dalam beberapa modalitas, dan pengambilan keputusan berdasarkan berbagai fitur yang datang dari berbagai sinyal sosial dengan berbagai modalitas.
|
Thesis (S2), Magang Riset, TA (S1) |
| 2 |
Kolmogorov-Arnold Networks & its applications
|
–
|
Thesis (S2), Magang Riset, TA (S1) | |
| 3 |
Pengembangan Sistem Terpadu Asesmen Psikologis Berbasis Multimodal Social Signal Processing dan Psikoedukasi pada Kader di Kabupaten Sukabumi
|
Topik ini berfokus pada pengembangan sistem asesmen psikologis berbasis multimodal social signal processing yang mengintegrasikan data kuesioner, ekspresi wajah, dan karakteristik suara untuk meningkatkan akurasi identifikasi kondisi psikologis. Sistem ini juga dilengkapi dengan fitur psikoedukasi mengenai psychological first aid, psikologi perkembangan, dan resiliensi keluarga. Fitur psikoedukasi ini disajikan dalam bentuk infografis dan gamifikasi sebagai media pembelajaran untuk mendukung peningkatan kapasitas kader dalam melakukan pendampingan pada masyarakat. Sistem dapat dikembangkan dalam bentuk mobile atau desktop app.
|
Magang Riset, TA (S1), Thesis (S2) | |
| Muhammad Febrian Rachmadi | 1 |
Lung nodule classification untuk screening lung cancer berdasarkan Lung-RADS
|
Bekerja sama dengan Departemen Radiologi FKUI untuk mengembangkan AI untuk radiologi , dimana lung nodules yang biasa ditemukan pada CT thorax pasien pre-cancer harus dianalisa jenisnya. Saat ini sudah terkumpul 400 pasien dengan lung nodules dari RSUI, dimana kedepannya akan dilakukan labeling dan analisis oleh dokter spesialis radiologi.
|
Magang Riset, TA (S1) |
| 2 |
ARRG untuk Chest X-Ray images
|
Bekerja sama dengan Departemen Radiologi FKUI untuk mengembangkan Automated Radiology Report Generation (ARRG) dari citra CT scans untuk berbagai kasus. Saat ini sudah terkumpul beberapa data CT scan dari RSUI.
|
Magang Riset, TA (S1) | |
| 3 |
Pendeteksian dan pengukuran fraktur wajah dari CT scan
|
Bekerja sama dengan Departemen Radiologi FKUI dan Bedah Plastik Rekonstruksi dan Estetik FKUI untuk melakukan pendeteksian, pengukuran, dan analisa fraktur wajah dari CT scan.
|
Magang Riset, TA (S1) | |
| 4 |
Klasifikasi gangguan urologi Bladder Outlet Obstruction (BOO) dan Detrusor Underactivity (DUA) berdasarkan video/citra cystoscopy
|
Bekerja sama dengan Departemen Urologi FKUI untuk mengembangkan AI untuk mengklasifikasikan 2 buah gangguan urologi pada pasien pria, yakni BOO dan DUA berdasarkan video/citra cystocopy dan fitir-fitur klinis lainnya (multimodal analysis).
|
Magang Riset, TA (S1) | |
| 5 |
Lung CT Thorax Analyzer untuk pasien Scleroderma-associated Interstitial Lung Disease (SSC-ILD)
|
Bekerja sama dengan Departemen Radiologi FKUI untuk penilaian Forced Vital Capacity (FVC) pada pasien SSC-ILD untuk analisa secara kuantitatif dan korelasi.
|
Magang Riset, TA (S1) | |
| 6 |
Embryo Analysis
|
Bekerja sama dengan Departemen Ob-Gin FKUI untuk mengembangkan model AI untuk menilai status kromosom embrio berdasarkan morfologi dan morfokinetik dari video time-lapse pada fertilisasi in vitro.
|
Magang Riset, TA (S1) | |
| Vektor Dewanto | 1 |
Discounting-free policy gradient reinforcement learning
|
Investigating RL approaches to do learning without discounting. We specifically focus on policy gradient methods.
|
Magang Riset, TA (S1), Thesis (S2) |
| 2 |
Task and motion planning using LLMs and OMPL for physical AI (robotics), eg, VLA (vision, language, action) models
|
Given a natural command, like “Tidy up this table”, a robot then plan the high level actions all the way down to low level motor commands.
We will leverage open-weight LLMs, classic motion planners OMPL, and physics engine Mujoco. |
Magang Riset, TA (S1), Thesis (S2) | |
| 3 |
Distributional reinforcement learning
|
One of newest approaches in RL is to target the distribution of cummulative rewards, rather than their expectation. Here, we will explore the ideas of distribution dynamics programming, then bring them to RL settings. We will begin with small and nice syntetic problems before deadling with complex real world ones.
|
Magang Riset, TA (S1), Thesis (S2) | |
| 4 |
Reinforcement Learning applications, including to robotics as in robot learning
|
RL is one of promising approaches to sequencial decision making under uncertainty. The applications of RL include dynamics pricing, inventory mangement, public heath policy making, robotics to name a few. Particularly in robotics, RL becomes the backbone for robot learning, eg learning to walk by trial and error in simulation.
|
Magang Riset, TA (S1), Thesis (S2) | |
| Ikhsanul Habibie | 1 |
AI-based 3D motion generation meets classical algorithms
|
This topic aims to explore how to combine AI-based models with classical motion generation ideas to synthesize realistic human movements from data with applications in animation, robotics, virtual reality, and AAA-rated video game. The topic focuses on generating smooth, natural, and controllable motions while reducing the need for expensive and tedious manual animation processes.
|
TA (S1), Thesis (S2), Magang Riset |
| 2 |
Full-body animatable 3D virtual human from a single camera
|
This research investigates how a complete animatable 3D virtual human can be reconstructed using only a single camera input. The study aims to generate realistic digital humans with accurate body shape, appearance, and motion capabilities.
|
Magang Riset, TA (S1) | |
| 3 |
Global 3D human pose estimation using few cameras
|
This topic aims to develop a novel 3D pose estimation method using few cameras focuses on recovering accurate full-body human poses in 3D space with minimal camera setups. The research combines multi-view geometry, deep learning, and pose estimation algorithms to reduce hardware complexity while maintaining reliable 3D pose prediction.
|
Magang Riset, TA (S1) | |
| 4 |
RL-based physics-aware and environment-aware 3D character control
|
This research investigates the use of reinforcement learning (RL) to develop intelligent 3D character control systems that can understand both physical dynamics and environmental conditions. By integrating physics simulation with environment-aware decision making, virtual characters can generate realistic, adaptive, and stable movements while interacting naturally with complex surroundings.
|
Magang Riset, TA (S1), Thesis (S2) | |
| 5 |
Markerless 3D Motion Capture for Clinical Gait Analysis
|
This research investigates camera-based motion capture systems that eliminate the need for wearable markers while still providing clinically accurate gait and movement analysis. It is highly relevant for rehabilitation centers, sports medicine, and elderly mobility assessment.
|
Magang Riset, TA (S1), Thesis (S2) | |
| Laksmita Rahadianti&Aruni Yasmin Azizah | 1 |
Spatial Data Processing for Environmental Monitoring
|
Environmental monitoring in Indonesia, especially in areas that are difficult to access, can be carried out using remote sensing imagery. From satellite images, spatial information can be obtained to support the analysis and decision makin, such as: Additionally, estimation from satellite imagery must first go through data processing to make it suitable for inference using various image processing techniques such as super-resolution, restoration, and inpainting. |
Magang RisetTA (S1) |
| 2 |
Indonesian Climate Model
|
Global climate change, which is becoming increasingly extreme, directly affects various sectors of life such as agriculture, food, and the environment. Climate models such as Global Climate Models (GCMs) are often used as tools to project climate conditions and their impacts. However, these models have limitations in terms of coarse resolution, making them insufficient as a basis for policymaking. Regional Climate Models (RCMs) are downscaling methods that can transform climate data into higher-resolution outputs, but they require high computational costs.
Deep learning methods can be utilized for estimating and projecting climate models, enabling high-resolution climate impact projections. This can be used to assess impacts across various sectors in Indonesia and serve as a foundation for policymaking. Some possible modeling approaches include: – Translation from low resolution to low resolution using different methods along with variable reduction/selection – Translation from low resolution to high resolution (without pre-processing) using different methods along with variable reduction/selection – And others. |
TA (S1)Thesis (S2) | |
| 3 |
Crop Analysis from Multimodal Remote Sensing Imagery
|
In collaboration with Departemen Geografi UGM – Crop detection and analysis from various satelite imagery. We will try some classical methods and also more advanced deep learning models to benchmark.
|
TA (S1)Thesis (S2) | |
| 4 |
Deepfake Detection
|
Nowadays, there are many deepfake techniques that can be used to impersonate others or misuse someone’s identity. This trend has accelerated with the rapid development of generative models and related technologies. As an initial study, this topic will explore various characteristics of deepfake images and evaluate and compare several existing detection techniques.
|
TA (S1)Magang Riset | |
| Adila Alfa Krisnadhi | 1 |
Large Concept Model for Legal Reasoning
|
–
|
Magang RisetTA (S1)Thesis (S2) |