RM新时代官方网站-首页

2023 NeurIPS（Neural Information Processing Systems）

One of the international AI academic conference NeurIPS （Neural Information Processing System announced the results of papers received in 2023. The paper “Hyperbolic Space with Hierarchical Margin Boosts Fine-Grained Learning from Coarse Labels” from AInnovation was successfully received by NeurIPS 2023. According to the statistics of NeurIPS official websites, 12343 valid papers were received in 2023 with an acceptance rate of 26.1%.

CVPR 2021 (International Conference on computer vision and pattern recognition)

The top global computer vision conference CVPR (IEEE Conference on computer vision and pattern recognition) announced the results of papers received in 2021. The paper "Zero-shot Instance Segmentation" from AInnovation was successfully received by CVPR 2021. According to the statistics of CVPR official website, more than 7500 valid papers were received in 2021, and finally 1663 papers were received, with an acceptance rate of 27.3%.

ECCV 2020 (European International Conference on computer vision)

AInnovtion’s research paper "Prototype Rectification for Few-Shot Learning" is accepted as an oral paper by the global computer vision summit ECCV 2020, with a selection rate of only 2%. This paper points out that the bottleneck of few-shot learning lies in the bias caused by data scarcity, mainly including intra class bias and cross class bias, and proposes corresponding methods to reduce the two biases.

2019 Design Automation Conference

AInnovation published a paper "Efficient GPU NVRAM Persistence with Helper Warps", and proposed a method to improve GPU performance and break through the bottleneck of computing power in the AI era by using NVRAM storage transaction processing system on GPU.

2019 ICCV International Conference on Computer Vision

AInnovation published the paper "HM-NAS: Efficient Neural Architecture Search via Hierarchical Masking" and won the best nomination in workshop. The paper proposes the search method of using hierarchical mask to make the model training faster, model parameters less, and network structure more optimal, thus requring lower computational power. It only needs 1-2 GPUs for one day training to complete the training generation of the model.

Top International Magazines《IEEE Transactions on Neural Networks and Learning Systems》

AInnovation published a paper "Deep Inception-Residual Laplacian Pyramid Networks for Accurate Single-Image Super-Resolution", in which the team proposed a new image super-resolution algorithm that can improve the accuracy of image reconstruction more quickly, and has very high industrial practical value in the fields of industrial product defect detection, precise measurement, commodity recognition, etc.