Neural Information Processing Systems (NeurIPS), one of the most prestigious research conferences in AI and machine learning technologies, have selected seven research papers from Microsoft for the oral presentations that will be held in the virtual week of the conference, from the 6th to 8th December 2022.
More than 150 researchers from Microsoft participated in the research conference, and 122 of their research papers were accepted. Out of the 122 accepted research papers, 7 of them were selected for oral presentations.
The titles of the seven research papers selected for oral presentations are as follows:
- Uni[MASK]: Unified Inference In Sequential Decision Problems
- Extreme Compression For Pre-Trained Transformers Made Simple and Efficient
- On the Complexity of Adversarial Decision Making
- Learning (Very) Simple Generative Models Is Hard
- Censored Quantile Regression Neural Networks For Distribution-Free Survival Analysis
- Maximum Class Separation As Inductive Bias In One Matrix
- K-LITE: Learning Transferable Visual Models With External Knowledge
Besides the above seven papers, two other research papers from Microsoft received the Outstanding Paper Awards in NeurIPS 2022. Gradient Estimation With Discrete Stein Operators is one of those papers that talks about a gradient estimator that achieves lower variance than the state-of-the-art estimators, potentially improving problem-solving in machine learning.
While the other paper, A Neural Corpus Indexer For Document Retrieval, tells about a deep neural network that unifies training and indexing stages to improve the recall performance of traditional document retrieval methods significantly.