Junyi An

Zhou Wu, Junyi An, Baile Xu et al.

Recently, the superior performance of Transformers has made them a more robust and scalable solution for sequence modeling than traditional recurrent neural networks (RNNs). However, the effectiveness of Transformer in capturing long-term dependencies is primarily attributed to their comprehensive pair-modeling process rather than inherent inductive biases toward sequence semantics. In this study, we explore the capabilities of pure RNNs and reassess their long-term learning mechanisms. Inspired by the physics energy transition models that track energy changes over time, we propose a effective recurrent structure called the``Physics-inspired Energy Transition Neural Network" (PETNN). We demonstrate that PETNN's memory mechanism effectively stores information over long-term dependencies. Experimental results indicate that PETNN outperforms transformer-based methods across various sequence tasks. Furthermore, owing to its recurrent nature, PETNN exhibits significantly lower complexity. Our study presents an optimal foundational recurrent architecture and highlights the potential for developing effective recurrent neural networks in fields currently dominated by Transformer.

Liang Pan, Tong Wu, Zhongang Cai et al.

As real-scanned point clouds are mostly partial due to occlusions and viewpoints, reconstructing complete 3D shapes based on incomplete observations becomes a fundamental problem for computer vision. With a single incomplete point cloud, it becomes the partial point cloud completion problem. Given multiple different observations, 3D reconstruction can be addressed by performing partial-to-partial point cloud registration. Recently, a large-scale Multi-View Partial (MVP) point cloud dataset has been released, which consists of over 100,000 high-quality virtual-scanned partial point clouds. Based on the MVP dataset, this paper reports methods and results in the Multi-View Partial Point Cloud Challenge 2021 on Completion and Registration. In total, 128 participants registered for the competition, and 31 teams made valid submissions. The top-ranked solutions will be analyzed, and then we will discuss future research directions.

Junyi An, Fengshan Liu, Jian Zhao et al.

Convolutional neural network (CNN) is a class of artificial neural networks widely used in computer vision tasks. Most CNNs achieve excellent performance by stacking certain types of basic units. In addition to increasing the depth and width of the network, designing more effective basic units has become an important research topic. Inspired by the elastic collision model in physics, we present a general structure which can be integrated into the existing CNNs to improve their performance. We term it the "Inter-layer Collision" (IC) structure. Compared to the traditional convolution structure, the IC structure introduces nonlinearity and feature recalibration in the linear convolution operation, which can capture more fine-grained features. In addition, a new training method, namely weak logit distillation (WLD), is proposed to speed up the training of IC networks by extracting knowledge from pre-trained basic models. In the ImageNet experiment, we integrate the IC structure into ResNet-50 and reduce the top-1 error from 22.38% to 21.75%, which also catches up the top-1 error of ResNet-100 (21.75%) with nearly half of FLOPs.

Junyi An, Fengshan Liu, Jian Zhao et al.

As a popular machine learning method, neural networks can be used to solve many complex tasks. Their strong generalization ability comes from the representation ability of the basic neuron model. The most popular neuron is the MP neuron, which uses a linear transformation and a non-linear activation function to process the input successively. Inspired by the elastic collision model in physics, we propose a new neuron model that can represent more complex distributions. We term it Inter-layer collision (IC) neuron. The IC neuron divides the input space into multiple subspaces used to represent different linear transformations. This operation enhanced non-linear representation ability and emphasizes some useful input features for the given task. We build the IC networks by integrating the IC neurons into the fully-connected (FC), convolutional, and recurrent structures. The IC networks outperform the traditional networks in a wide range of experiments. We believe that the IC neuron can be a basic unit to build network structures.

Junyi An, Fengshan Liu, Jian Zhao et al.

Neural networks have been widely used, and most networks achieve excellent performance by stacking certain types of basic units. Compared to increasing the depth and width of the network, designing more effective basic units has become an important research topic. Inspired by the elastic collision model in physics, we present a universal structure that could be integrated into the existing network structures to speed up the training process and increase their generalization abilities. We term this structure the "Inter-layer Collision" (IC) structure. We built two kinds of basic computational units (IC layer and IC block) that compose the convolutional neural networks (CNNs) by combining the IC structure with the convolution operation. Compared to traditional convolutions, both of the proposed computational units have a stronger non-linear representation ability and can filter features useful for a given task. Using these computational units to build networks, we bring significant improvements in performance for existing state-of-the-art CNNs. On the imagenet experiment, we integrate the IC block into ResNet-50 and reduce the top-1 error from 22.85% to 21.49%, which also exceeds the top-1 error of ResNet-100 (21.75%).