Neuro-symbolic Deep Learning for Logical Inference

Description

Abilities to do logical inference, i.e. rule-based generalization, distinguish human intelligence from contemporary artificial intelligence (AI) implemented by deep learning models. To bridge this gap, studies about neuro-symbolic deep learning approach have been conducted. In this internship, we aim to digest existing works about the neuro-symbolic approach by reading papers and analyzing their implementation codes. To go further, we implement our own ideas and tackle challenging systematic generalization problems.

Candidate Qualifications

• Strong and persistent attitude for the research (required)
• Basic python skills (required)
• PyTorch experience (optional)

Expected Internship Period

• Minimum three months

Contact

• Segwang Kim (ksk5693@snu.ac.kr)

End-to-end text-to-speech (Closed)

Description

Investigating end-to-end learning approaches to building a text-to-speech system that converts natural language text into human speech. The internship includes research, implementation, and experiment for target tasks above.

Candidate Qualifications

• Strong attitude for the investigation
• Basic python skills (required)
• Understanding of recent algorithm in deep learning literature (i.e. Tacotron from google)
• Experience of implementing an algorithm using PyTorch or Tensorflow

Expected Internship Period

• Minimum four months - excluding period for the preliminary study (Pytorch, base ML algorithms)

Contact

• Yoonhyung Lee (cpi1234@snu.ac.kr) (Completed)

Adversarial text generation (Closed)

Description

Adversarial Text generation aims at generating text with Generative adversarial networks (GANs) which can alleviate those drawbacks of MLE-based models such as exposure bias. You will learn the procedure of deep learning research which includes design and implementation of deep learning models. In addition, you will be involved in the research in progress.

Candidate Qualifications

• Strong attitude for the investigation
• Basic python skills (required)
• Familiar with fundamental deep learning models (FNNs, CNNs, RNNs and etc)
• Experience of implementing algorithms using deep learning open-source software library (Optional)

Expected Internship Period

• Minimum three months - excluding period for the preliminary study (deep learning basics, library)

Contact

• Yanghoon Kim (ad26kr@snu.ac.kr) (Completed)

Recommendation systems using deep learning (Closed)

Description

Investigating deep learning algorithms or approaches to improve recommendation systems that seeks to predict the preference of users. The internship includes research, implementation, and experiment for target tasks above.

Candidate Qualifications

• Strong attitude for the investigation
• Basic python skills (required)
• Experience of implementing an algorithm using Tensorflow or PyTorch (optional)

Expected Internship Period

• Minimum three months - excluding period for the preliminary study (Tensorflow, base ML algorithms)

Contact

• Byeongjin Choe (bjchoe@snu.ac.kr) (Completed)

Deep Learning-based Sequence to Sequence Systematic Generalization (Closed)

Description

Given a sequence of symbols, predicting the next or corresponding sequence of symbols is of interest since this sequence transduction problem can be found in many applications like translation. Recently neural network model "transformer" has achieved remarkable results in translation by allowing all pairwise interactions between symbols within input and target sequence. However, transformer leaves a question whether the model can compose outputs in systematical ways. The internship aims to improve existing models to generalize simple matching rules or algebraic ones in algorithmic tasks. The internship includes research, implementation, and experiment for target tasks above.

Candidate Qualifications

• Strong attitude for the investigation
• Basic python skills (required)
• Experience of implementing an algorithm using Tensorflow (optional)
• Experience of using high-level frameworks on Tensorflow (e.g. Tensor2Tensor, optional)

Expected Internship Period

• Minimum three months - excluding period for the preliminary study (Tensorflow, base ML algorithms)

Contact

• Segwang Kim (ksk5693@snu.ac.kr) (Completed)

Speech Emotion Recognition via a Representation Learning (Completed)

Description

Investigating algorithms (or approaches) to improve the performance of speech emotion recognition system that classifies speech signal into sets of emotion label such as "happy", "sad", "angry", and "neutral". The internship includes research, implementation, and experiment for target tasks above.

Candidate Qualifications

• Strong attitude for the investigation
• Understanding of recent algorithm in deep learning literature (i.e. BERT from google)
• Experience of implementing an algorithm using Tensorflow (optional)
• Familiar with speech signal processing (optional)

Expected Internship Period

• Minimum four months - excluding period for the preliminary study (Tensorflow, base ML algorithms)

Contact

• Seunghyun Yoon (mysmilesh@snu.ac.kr) (Completed)

• click to apply (Completed)