News & features
Enhancing your photos through artificial intelligence
The amount of visual data we accumulate around the world is mind boggling. However, not all the images are captured by high-end DSLR cameras, and very often they suffer from imperfections. It is of tremendous benefit to save those degraded…
High-Resolution Network: A universal neural architecture for visual recognition
Since AlexNet was invented in 2012, there has been rapid development in convolutional neural network architectures in computer vision. Representative architectures (Figure 1) include GoogleNet (2014), VGGNet (2014), ResNet (2015), and DenseNet (2016), which are developed initially from image classification.…
XGLUE: Expanding cross-lingual understanding and generation with tasks from real-world scenarios
| Nan Duan, Yaobo Liang, and Daniel Campos
What we can teach a model to do with natural language is dictated by the availability of data. Currently, we have a lot of labeled data for very few languages, making it difficult to train models to accomplish question answering,…
Earlier this year, a novel coronavirus rendered online teaching a norm. With “classroom” experiences poor, effective interactions between teachers and students difficult to maintain, and teaching quality easily affected by network conditions, online teaching has created many difficulties for teachers…
He was the first Chinese scholar to become a “complete” fellow at all the following organizations: ACM, AAAI, IEEE, AAAS and IAPR. He was the first Chinese scholar to serve as the program chair of AAAI. He was elected as…
Researchers in Microsoft Research Asia lab launch COVID Insights website
As the COVID-19 pandemic continues to impact communities around the world, researchers everywhere are fighting hard in the hopes of gradually understanding the mystery of the COVID-19 virus through the most advanced technologies available. Based on their expertise and research…
In the news | VentureBeat
Microsoft’s CodeBERT ingests public GitHub repositories to help you find code
Large pretrained language models have improved the state-of-the-art on a range of natural language processing tasks, chiefly because they’re able to learn contextual representations from text without supervision. In a preprint paper, a team of researchers at Microsoft Research Asia used…
Finding the best learning targets automatically: Fully Parameterized Quantile Function for distributional RL
| Li Zhao
Reinforcement learning has achieved great success in game scenarios, with RL agents beating human competitors in such games as Go and poker. Distributional reinforcement learning, in particular, has proven to be an effective approach for training an agent to maximize…
FastSpeech: New text-to-speech model improves on speed, accuracy, and controllability
| Xu Tan
Text to speech (TTS) has attracted a lot of attention recently due to advancements in deep learning. Neural network-based TTS models (such as Tacotron 2, DeepVoice 3 and Transformer TTS) have outperformed conventional concatenative and statistical parametric approaches in terms…