Oleksiy is a machine learning and NLP researcher. For the past 10+ years, he worked on grammatical error correction, and AutoML platform, content recommendation, online advertising, and other products.
Oleksiy teaches a Deep Learning for NLP course at Lviv Polytechnic National University.
Transfer learning for speech processing
Data Science and Machine Learning
Session Language |Ukraine
Transfer learning drives many recent advances in deep learning. Models like BERT, GPT-3, CLIP, and Codex are first pre-trained on large unlabeled datasets. Then these models could be fine-tuned on many downstream tasks with comparatively little labeled data. This approach became mainstream in textual NLP and computer vision.But what about speech processing?This talk will explore how transfer learning reduces the need for labeled data in applications like automatic speech recognition and text-to-speech synthesis. We'll explore emerging areas like direct speech-to-speech translation and textless NLP.