Title: Attention Guided Graph Convolutional Networks for Relation Extraction
Speaker: Guogen Tang
Dependency trees convey rich structural information that is proven useful for extracting relations among entities in text. However, how to effectively make use of relevant information while ignoring irrelevant information from the dependency trees remains a challenging research question. Existing approaches employing rule based hard-pruning strategies for selecting relevant partial dependency structures may not always yield optimal results. In this work, we propose Attention Guided Graph Convolutional Networks (AGGCNs), a novel model which directly takes full dependency trees as inputs. Our model can be understood as a soft-pruning approach that automatically learns how to selectively attend to the relevant sub-structures useful for the relation extraction task. Extensive results on various tasks including cross-sentence n-ary relation extraction and large-scale sentence-level relation extraction show that our model is able to better leverage the structural information of the full dependency trees, giving significantly better results than previous approaches.
Supervisor: Biao Wang
Title: You Only Look Once: Unified, Real-Time Object Detection
Speaker: Li Gou
Based on deep learning, the object detection algorithm is divided into two parts,two-stage and one-stage,where YOLO is a milestone about them. Prior work of YOLO on object detection repurposes classififiers to perform detection.Two-stage algorithm such as R-CNN,Fast R-CNN, Faster R-CNN has advantages in accuracy, but they do not good at speed.YOLO frame object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance. We analyzed the advantages and disadvantages of YOLO v1,v2 and v3.Starting from YOLO v1 , YOLO is to detect by dividing cells, but the number of divisions is different.It just need a loss function, just pay attention to the input and the output.YOLOv2 is proposed after making improvements on the basis of v1. It was inspired by faster RCNN and introduced the anchor. At the same time, the k-means method is used to discuss the number of anchors. And modified the network structure, removed the fully connected layer, and changed it to a fully convolutional structure.In YOLO3,through the integration of a variety of advanced methods, the shortcomings of the YOLO series (fast, not good at detecting small objects) are all filled.On PASCAL VOC2007，YOLOv1 scores 63.4 mAP and 45 FPS, YOLOv2 scores 77.8 mAP and 67 FPS.On the 156 classes not in COCO, YOLO V2 gets 16.0 mAP. It achieves 57.9 AP50 in 51 ms on a Titan X, compared to 57.5 AP50 in 198 ms by RetinaNet, similar performance but 3.8× faster.The series of YOLO has been updated to the v5,we will continue looking for ways to bring different sources and structures of data together to make stronger models of visual world.
Supervisor: Guogen Tang