Fei Chiang

Nishttha Sharma, Fei Chiang

Understanding data change is critical towards understanding trends, normal vs. abnormal behaviours, recognizing patterns, and the causes of change. Existing database systems have limited support for change management, relying on statistics, triggers, and constraints. Data quality rules model sequential changes along a restricted set of attributes, quantify change among unordered tuples, and have limited ability to model the context under which attribute changes occur. In this paper, we introduce Change Rules (CRs) that quantify the sequential changes among ordered tuples in both the antecedent and consequent attributes. CRs aim to address the limitations of existing declarative dependencies to support trend analysis and causal relationships that trigger change among attributes. We propose CR-Miner, an automated algorithm for CR discovery that generates candidate change intervals in a level-wise manner. Experimental results show that CR-Miner achieves an average runtime improvement of 40-50% over existing baselines.

Shervin Manzuri Shalmani, Fei Chiang, Rong Zheng

Modern neural networks are powerful predictive models. However, when it comes to recognizing that they may be wrong about their predictions, they perform poorly. For example, for one of the most common activation functions, the ReLU and its variants, even a well-calibrated model can produce incorrect but high confidence predictions. In the related task of action recognition, most current classification methods are based on clip-level classifiers that densely sample a given video for non-overlapping, same-sized clips and aggregate the results using an aggregation function - typically averaging - to achieve video level predictions. While this approach has shown to be effective, it is sub-optimal in recognition accuracy and has a high computational overhead. To mitigate both these issues, we propose the confidence distillation framework to teach a representation of uncertainty of the teacher to the student sampler and divide the task of full video prediction between the student and the teacher models. We conduct extensive experiments on three action recognition datasets and demonstrate that our framework achieves significant improvements in action recognition accuracy (up to 20%) and computational efficiency (more than 40%).