Thesis supervisor: Bálint Gyires-Tóth
Location of studies (in Hungarian): Távközlési és Médiainformatikai Tanszék Abbreviation of location of studies: TMIT
Description of the research topic:
Due to the huge increase in the amount of available data, the rise of high-performance GPUs and the novel results in neural networks, deep learning has become a focused research topic among machine learning techniques in last years. Deep learning has revolutionized many applied research areas, like speech recognition and synthesis, computer vision, natural language processing and understanding. The modeling performance of deep learning systems significantly surpasses earlier machine learning methods due to their ability to extract and learn abstractions of a huge amount of input data. State-of-the-art Natural Language Processing (NLP) solutions typically require massive amount of data and computational capacity. However, specific domains and low resource languages don’t have large corpora; furthermore, the computational capacity is limited in case of most research groups. Additionally, distributed representations and deep learning based NLP models are still not understood, and mostly their behavior cannot be explained. Thus, the long-term trend of using bigger, uncomprehended models and larger datasets in Natural Language Processing is obscure and might be unsustainable in many future cases.
The goal of the current research is to create novel algorithms that support sustainable deep learning based NLP in terms of dataset size, computational resources and/or explainability. The research should investigate inter alia the utilization of transfer learning; zero-, one- and few-shot learning; unsupervised learning, meta-learning; and/or deep reinforcement learning in Natural Language Processing solutions. The work may also involve developing novel deep learning approaches in NLP to achieve sustainability. The effectiveness of the elaborated method must be proven, at least in one application scenario. Such an application scenario can be (1) speech technologies, (2) neural machine translation, (3) question answering, etc.
The research can be performed in English and Hungarian. For training the models public and private databases and high-performance GPUs are available.
The possible research tasks of the Ph.D. candidatee are the following:
- Overview of the related scientific papers, including basic deep neural networks elements and novel results in Natural Language Processing.
- Design and implement a baseline and existing state-of-the-art NLP solutions for a benchmark.
- Elaborate novel deep learning based NLP solutions that address sustainablility in terms of dataset size, computational resources and/or explainable machine learning.
- Demonstrate the effectiveness of the results at least in one application scenario.
- Objective and subjective evaluation.
Further requirements: English or Hungarian
Number of students who can be accepted: 1
Deadline for application: 2020-06-15
2024. IV. 17. ODT ülés Az ODT következő ülésére 2024. június 14-én, pénteken 10.00 órakor kerül sor a Semmelweis Egyetem Szenátusi termében (Bp. Üllői út 26. I. emelet).
Login
