Job offer in Ireland Deploying Computer Vision Tasks in the Wild Driven by Meta-learning at Anyvision
Thanks to the resurgence of deep learning, in recent years, computer vision has achieved stunning progress and been applied to many real-world applications, e.g., face recognition and person re-identification in surveillance cameras. However, those approaches heavily rely on a significant amount of labeled data and extensive parameter tuning. When deploying computer vision tasks in the real-world, especially under unseen scenarios, the pre-trained models generally perform poorly and can’t quickly generalize with few shots. Similar to meta-level construct in biology, meta-learning (a.k.a., learning to learn) aims to acquire a meta-level knowledge across tasks and shifts its inductive bias via fast parameterization for the rapid generalization. It has attracted great attention recently and showed promising results on classical computer vision tasks [1-6].
The goal of this Ph.D. project is to deploy general computer vision tasks in the unseen scenarios with minimum efforts. The main work directions as below:
- Investigating to which extent deep architectures can be used for efficient meta-learning
- Designing a better meta-learning approach for deploying general computer vision tasks in the unseen scenarios
The project is under collaboration with Anyvision where the latter provides adequate funding, extra computing resources and opportunity to work with a skilled and vibrant team of researchers and engineers. AnyVision is a rapidly expanding company in advanced real-time object recognition from big data, providing world-leading solutions in authentication, surveillance, security and social settings. With offices in Tel Aviv and New York, sales staff all over the world, the research team of Anyvision is based in Belfast, UK.
- Master’s degree or expected to obtain it soon in computer science, engineering, mathematics, or related fields (outstanding student with prestigious Bachelor’s degree can also be considered)
- Solid mathematics knowledge and programming skills
- Fluent in English (for an international applicant, the minimum requirement is IELTS 6.0 with a minimum of 5.5 in all four elements)
- Prior knowledge in the areas of computer vision and machine learning (especially deep learning)
- Real-world project experience or working experience in the areas of computer vision and machine learning (especially deep learning) is plus
- Full tuition fee and stipend, plus sponsor company top-up
- Sufficient hardware resource
- Working within a vibrant research and engineering team
The application shall include the following documents to firstname.lastname@example.org:
- Curriculum vitae with a focus on the qualifications of interest listed above
- Official Transcripts for Bachelor and Master study
- Cover Letter with self-introduction and the past education, research, and project/working experiences focusing on the related subjects to this position. (max 2 pages)
 Zhenguo Li, Fengwei Zhou, Fei Chen, Hang Li. Meta-SGD: Learning to Learn Quickly for Few-Shot Learning. arXiv 2017.
 Flood Sung, Zhang L, Xiang T, Hospedales T, et al. Learning to Learn: Meta-Critic Networks for Sample Efficient Learning. arXiv, 2017.
 Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. ICLR 2017
 Tsendsuren Munkhdalai and Hong Yu. Meta Networks. arXiv, 2017.
 Janghoon Choi, Junseok Kwon, Kyoung Mu Lee. Deep Meta Learning for Real-Time Visual Tracking based on Target-Specific Feature Space. arXiv 2017.
 Eunbyung Park, Alexander C. Berg. Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers. arXiv 2018.