A5091175785- Topic Modeling
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Neural Networks and Applications
- Domain Adaptation and Few-Shot Learning
- Speech Recognition and Synthesis
- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Text Readability and Simplification
- Machine Learning in Bioinformatics
- Speech and dialogue systems
- Music and Audio Processing
- Total Knee Arthroplasty Outcomes
- Machine Learning and Data Classification
- Protein Structure and Dynamics
- Reinforcement Learning in Robotics
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Computational Drug Discovery Methods
- Machine Learning in Materials Science
- Generative Adversarial Networks and Image Synthesis
- Machine Learning and Algorithms
- Advanced Image and Video Retrieval Techniques
- Stochastic Gradient Optimization Techniques
- Speech and Audio Processing
New York University
2016-2025
Courant Institute of Mathematical Sciences
2016-2025
Mercer University
2024
Canadian Institute for Advanced Research
2013-2023
University of Washington
2020-2023
Carnegie Mellon University
2020-2023
Korea Advanced Institute of Science and Technology
2023
Johns Hopkins University
2023
Shanghai Jiao Tong University
2023
Massachusetts Institute of Technology
2023
Neural machine translation is a relatively new approach to statistical based purely on neural networks.The models often consist of an encoder and decoder.The extracts fixed-length representation from variable-length input sentence, the decoder generates correct this representation.In paper, we focus analyzing properties using two models; RNN Encoder-Decoder newly proposed gated recursive convolutional network.We show that performs well short sentences without unknown words, but its...
Recurrent sequence generators conditioned on input data through an attention mechanism have recently shown very good performance a range of tasks in- cluding machine translation, handwriting synthesis and image caption gen- eration. We extend the attention-mechanism with features needed for speech recognition. show that while adaptation model used translation in reaches competitive 18.7% phoneme error rate (PER) TIMIT recognition task, it can only be applied to utterances which are roughly...
Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We how can train this a deterministic manner using standard backpropagation techniques stochastically maximizing variational lower bound. also show through visualization is able learn fix its gaze on salient objects while generating corresponding words output sequence. validate use with state-of-the-art performance three...
Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. Since its introduction, it has been one of the most used CPU GPU compilers - especially in machine learning community shown steady performance improvements. being actively continuously developed since 2008, multiple frameworks have built on top produce many state-of-the-art models. The present article structured as follows. Section I provides an...
Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application video description. However, while images are static, working with videos requires modeling dynamic temporal structure and then properly integrating that information into a natural language model. In this context, we propose an approach successfully takes account both local global to produce descriptions. First, our incorporates spatial 3-D convolutional network...
Sébastien Jean, Kyunghyun Cho, Roland Memisevic, Yoshua Bengio. Proceedings of the 53rd Annual Meeting Association for Computational Linguistics and 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.
Neural machine translation is a relatively new approach to statistical based purely on neural networks. The models often consist of an encoder and decoder. extracts fixed-length representation from variable-length input sentence, the decoder generates correct this representation. In paper, we focus analyzing properties using two models; RNN Encoder--Decoder newly proposed gated recursive convolutional network. We show that performs well short sentences without unknown words, but its...
Abstract The rapid increase in the number of proteins sequence databases and diversity their functions challenge computational approaches for automated function prediction. Here, we introduce DeepFRI, a Graph Convolutional Network predicting protein by leveraging features extracted from language model structures. It outperforms current leading methods sequence-based Neural Networks scales to size repositories. Augmenting training set experimental structures with homology models allows us...
Recent work on end-to-end neural network-based architectures for machine translation has shown promising results En-Fr and En-De translation. Arguably, one of the major factors behind this success been availability high quality parallel corpora. In work, we investigate how to leverage abundant monolingual corpora Compared a phrase-based hierarchical baseline, obtain up $1.96$ BLEU improvement low-resource language pair Turkish-English, $1.59$ focused domain task Chinese-English chat...
We present a deep convolutional neural network for breast cancer screening exam classification, trained, and evaluated on over 200000 exams (over 1000000 images). Our achieves an AUC of 0.895 in predicting the presence breast, when tested population. attribute high accuracy to few technical advances. 1) network's novel two-stage architecture training procedure, which allows us use high-capacity patch-level learn from pixel-level labels alongside learning macroscopic breast-level labels. 2) A...
We propose multi-way, multilingual neural machine translation.The proposed approach enables a single translation model to translate between multiple languages, with number of parameters that grows only linearly the languages.This is made possible by having attention mechanism shared across all language pairs.We train multiway, on ten pairs from WMT'15 simultaneously and observe clear performance improvements over models trained one pair.In particular, we significantly improves quality...
Unsupervised methods for learning distributed representations of words are ubiquitous in today's NLP research, but far less is known about the best ways to learn phrase or sentence from unlabelled data. This paper a systematic comparison models that such representations. We find optimal approach depends critically on intended application. Deeper, more complex preferable be used supervised systems, shallow log-linear work building representation spaces can decoded with simple spatial distance...
We study the complexity of functions computable by deep feedforward neural networks with piecewise linear activations in terms symmetries and number regions that they have. Deep are able to sequentially map portions each layer's input-space same output. In this way, models compute react equally complicated patterns different inputs. The compositional structure these enables them re-use pieces computation exponentially often network's depth. This paper investigates such maps contributes new...
Neural machine translation is a recently proposed approach to translation. Unlike the traditional statistical translation, neural aims at building single network that can be jointly tuned maximize performance. The models for often belong family of encoder-decoders and consists an encoder encodes source sentence into fixed-length vector from which decoder generates In this paper, we conjecture use bottleneck in improving performance basic encoder-decoder architecture, propose extend by...
We introduce a convolutional recurrent neural network (CRNN) for music tagging. CRNNs take advantage of networks (CNNs) local feature extraction and temporal summarisation the extracted features. compare CRNN with three CNN structures that have been used tagging while controlling number parameters respect to their performance training time per sample. Overall, we found show strong parameter time, indicating effectiveness its hybrid structure in summarisation.
Whereas deep neural networks were first mostly used for classification tasks, they are rapidly expanding in the realm of structured output problems, where observed target is composed multiple random variables that have a rich joint distribution, given input. We focus this paper on case input also has structure and structures somehow related. describe systems learn to attend different places input, each element output, variety tasks: machine translation, image caption generation, video clip...
We propose a conditional non-autoregressive neural sequence model based on iterative refinement. The proposed is designed the principles of latent variable models and denoising autoencoders, generally applicable to any generation task. extensively evaluate machine translation (En-De En-Ro) image caption generation, observe that it significantly speeds up decoding while maintaining quality comparable autoregressive counterpart.
Jonas Pfeiffer, Andreas Rücklé, Clifton Poth, Aishwarya Kamath, Ivan Vulić, Sebastian Ruder, Kyunghyun Cho, Iryna Gurevych. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations. 2020.