Deep Learning
On Adversarial Mixup Resynthesis
Christopher Beckham, Sina Honari, Vikas Verma, Alex M. Lamb, Farnoosh Ghadiri, R Devon Hjelm, Yoshua Bengio, Chris Pal
The auto-encoder is a fundamental building block in unsupervised learning.
Park: An Open Platform for Learning-Augmented Computer Systems
Hongzi Mao, Parimarjan Negi, Akshay Narayan, Hanrui Wang, Jiacheng Yang, Haonan Wang, Ryan Marcus, ravichandra addanki, Mehrdad Khani Shirkoohi, Songtao He, Vikram Nathan, Frank Cangialosi, Shaileshh Venkatakrishnan, Wei-Hung Weng, Song Han, Tim Kraska, Dr.Mohammad Alizadeh
Neural Information Processing Systems http://nips.cc/
FastSpeech: Fast, Robust and Controllable Text to Speech
Yi Ren, Yangjun Ruan, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, Tie-Yan Liu
Prominent methods (e.g., Tacotron 2) usually first generate mel-spectrogram from text, and then synthesize speech from the mel-spectrogram using vocoder such as WaveNet. Compared with traditional concatenative and statistical parametric approaches, neural network based end-to-end models suffer from slow inference speed, and the synthesized speech is usually not robust (i.e., some words are skipped or repeated) and lack of con-trollability (voice speed or prosody control).
both R1 and R2 see the significance of these contributions, and R1 is correct: we have gotten a lot of interest in the
We would like to thank the reviewers for their thoughtful comments. However, R2 and R3 had some concerns. Based on initial interest, it seems that many groups will investigate these questions. We think there's value in making these points in the context of these newer techniques. There is also more nuance to our results that R3 is not giving us credit for.
we want to emphasize that the goal of S PECTRL
We thank the reviewers for their helpful suggestions, and will do our best to incorporate them into our paper. It is true that LSTMs can be used to solve RL problems with non-Markovian specifications. As we show in our experiments, reward shaping is crucial for learning complex tasks. For example, how does achieving a sub-goal count compared to violating a constraint? These challenges are exactly what our system is designed to solve.
Deep Multimodal Multilinear Fusion with High-order Polynomial Pooling
Ming Hou, Jiajia Tang, Jianhai Zhang, Wanzeng Kong, Qibin Zhao
Tensor-based multimodal fusion techniques have exhibited great predictive performance.
Compositional generalization through meta sequence-to-sequence learning
Once a person learns the meaning of the verb "to Facebook", she or he
Neural Taskonomy: Inferring the Similarity of Task-Derived Representations from Brain Activity
Aria Wang, Michael Tarr, Leila Wehbe
Using results across all 21 task representations, we constructed a "task graph" based on the spatial layout of well-predicted brain areas from each task.
Generative Models for Graph-Based Protein Design
John Ingraham, Vikas Garg, Regina Barzilay, Tommi Jaakkola
Engineered proteins offer the potential to solve many problems in biomedicine, energy, and materials science, but creating designs that succeed is difficult in practice.
Deep Learning without Weight Transport
Mohamed Akrout, Collin Wilson, Peter Humphreys, Timothy Lillicrap, Douglas B. Tweed
In a typical deep-learning network, some signals flow along a forward path through multiple layers of processing units from the input layer to the output, while other signals flow back from the output layer along a feedback path . Forward-path signals perform inference (e.g. they try to infer what objects are