Search

Structure pruning is an effective method to compress and accelerate neural networks. While filter and channel pruning are preferable to other structure pruning methods in terms of realistic acceleration and hardware compatibility, pruning methods with a finer granularity, such as intra-channel pruning, are expected to be capable of yielding more compact and computationally efficient networks. Typical intra-channel pruning methods utilize a static and hand-crafted pruning granularity due to a large search space, which leaves room for improvement in their pruning performance. In this work, we introduce a novel structure pruning method, termed as dynamic structure pruning, to identify optimal pruning granularities for intra-channel pruning. In contrast to existing intra-channel pruning methods, the proposed method automatically optimizes dynamic pruning granularities in each layer while training deep neural networks. To achieve this, we propose a differentiable group learning method designed to efficiently learn a pruning granularity based on gradient-based learning of filter groups. The experimental results show that dynamic structure pruning achieves state-of-the-art pruning performance and better realistic acceleration on a GPU compared with channel pruning. In particular, it reduces the FLOPs of ResNet50 by 71.85% without accuracy degradation on the ImageNet dataset. Our code is available at https://github.com/irishev/DSP.

One important strength of the microlensing method in detecting extrasolar planets is its high sensitivity to low-mass planets. However, it is often believed that microlensing detections of Earth-mass planets from ground-based observation would be difficult due to the limit set by finite-source effects. This view comes from the previous estimation of the planet detection probability based on the fractional deviation of planetary signals, but proper probability estimation requires to additionally consider the source brightness, which is directly related to the photometric precision. In this paper, we reevaluate the feasibility of low-mass planet detections considering photometric precision for different populations of source stars. From this, it is found that contribution of the improved photometric precision to the planetary signal of a giant-source event is big enough to compensate the decrease of the magnification excess caused by finite-source effects. As a result, we find that giant-source events are suitable targets for Earth-mass planet detections with significantly higher detection probability than events involved with source stars of smaller radii and predict that Earth-mass planets would be detected by prospective high-cadence surveys., Comment: 4 pages, 3 figures, ApJ accepted