Master Thesis – Enhanced Reasoning in Vision-Language Models Using Structured Visual Representations (AI)
Контакты
Формат работы
remote (только Sweden)/onsite
Тип работы
fulltime
Грейд
trainee
Английский
b2
Страна
Sweden
machine learningcomputer visionvideo analyticslorapeftvision language modelsspatial reasoningstructured visual representationsinstruction fine tuningtemporal reasoning
Вакансия из Hirify Global, списка международных tech-компаний Для мэтча и отклика нужен Plus
Пожаловаться
88
Крутая вакансия
развернуть
Чётко описанная исследовательская работа в известной технологической компании посвящена актуальной теме VLM и включает практическую настройку и оценку моделей; участвовать могут только студенты, связанные со шведским вузом.
Ограничение по вузу
Чёткая тема
Актуальная область
Известный работодатель
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Подключить Plus
Описание вакансии
Текст:
Hirify AI/Оригинал
TL;DR
Master Thesis – Enhanced Reasoning in Vision-Language Models Using Structured Visual Representations (AI): Investigating how structured visual metadata such as object locations, identities, trajectories, and velocity vectors can improve reasoning in Vision-Language Models with an accent on video understanding, spatial reasoning, and temporal reasoning. Focus on fine-tuning open-source VLMs with instruction fine-tuning and PEFT/LoRA, integrating intermediate visual representations, and evaluating interpretability and performance.
Location: Lund, Sweden; workplace type: on-site. The thesis is open only to students affiliated with a Swedish University/College directly or through an exchange program.
Company
hirify.global develops network technology and intelligent analytics solutions for video surveillance, access control, intercom, and audio systems.
What you will do
Investigate state-of-the-art methods for integrating structured visual information into Vision-Language Models.
Use surveillance metadata, including bounding boxes, object identities, tracking information, trajectories, and velocity vectors, and explore training data generation from in-house and public datasets.
Fine-tune one or more open-source VLMs using instruction fine-tuning and Parameter-Efficient Fine-Tuning, such as LoRA.
Explore how intermediate visual representations can be represented, aligned with visual input, and incorporated into the model's reasoning during training and inference.
Evaluate object-level, spatial, and temporal reasoning in video question answering and compare models trained with and without structured visual information.
Investigate the interpretability and limitations of the resulting approach.
Requirements
Current enrollment in a Master's program in Electronics Engineering, Mathematics, Computer Science, Biomedical Engineering, or Engineering Physics.
Strong interest in machine learning, Vision-Language Models, problem solving, and programming.
The thesis scope is for two Master's students, 30 hp each; applicants are encouraged to apply in pairs and identify their co-applicant.
Applicants must be affiliated with a Swedish University/College directly or through an exchange program.
Culture & Benefits
Full-time thesis worker position with a fixed-term arrangement.
Applications are accepted in Swedish and English.
Application materials must include a CV, cover letter, and University/College grade summary.
Background checks may apply to certain roles, with notice provided before action is taken.
Hiring process
Apply through the thesis proposal advert.
When applying as a pair, submit one application per student and identify the co-applicant.
Будьте осторожны
Если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого. Это мошенники. Обязательно жмите «Пожаловаться» или пишите в поддержку. Подробнее в гайде →