Activated Gradients for Deep Neural Networks.

Deep neural networks often suffer from poor performance or even training failure due to the ill-conditioned problem, the vanishing/exploding gradient problem, and the saddle point problem. In this article, a novel method by acting the gradient activation function (GAF) on the gradient is proposed to handle these challenges. Intuitively, the GAF enlarges the tiny gradients and restricts the large gradient. Theoretically, this article gives conditions that the GAF needs to meet and, on this basis, proves that the GAF alleviates the problems mentioned above. In addition, this article proves that the convergence rate of SGD with the GAF is faster than that without the GAF under some assumptions. Furthermore, experiments on CIFAR, ImageNet, and PASCAL visual object classes confirm the GAF's effectiveness. The experimental results also demonstrate that the proposed method is able to be adopted in various deep neural networks to improve their performance. The source code is publicly available at https://github.com/LongJin-lab/Activated-Gradients-for-Deep-Neural-Networks.

Clinical Outcomes of Aberration-Free All Surface Laser Ablation (ASLA) vs. Aberration-Free ASLA Assisted by Smart Pulse Technology in High Myopia: A One-Year Follow-Up Study.

PURPOSE: To compare the clinical outcomes of aberration-free all surface laser ablation (ASLA) with and without the use of smart pulse technology (SPT) in high myopia. METHODS: This study retrospectively analyzed 138 eyes (138 patients, only the right eye was selected) treated for high myopia (spherical equivalent ≥-6.00 diopters) using aberration-free ASLA (non-SPT group; 85 eyes) and aberration-free ASLA assisted by SPT (SPT group; 53 eyes). Examinations such as visual acuity, refraction, and haze were performed before the 12-month follow-up. Corneal epithelial healing time was assessed in the first postoperative day. Visual acuity and refraction examination were performed at 7 days and 1, 3, 6, and 12 months postoperatively. Corneal haze was evaluated in 1, 3, 6, and 12 months. Safety, efficacy, and corneal wavefront aberrations were assessed 12 months after the treatment. RESULTS: At 12 months postoperatively, 60% versus 40% of eyes achieved 20/16 Snellen lines or better, and 92% versus 82% of eyes achieved 20/20 Snellen lines or better visual acuity in the SPT and the non-SPT groups, respectively. The average postoperative epithelial healing time was 3.75 ± 1.00 days in the SPT group and 3.73 ± 1.30 days in the non-SPT group (P ≥ 0.05). The safety and the efficacy index of the SPT group were better than those of the non-SPT group in the follow-ups. The attempted spherical equivalent before the surgery and the achieved spherical equivalent at 12 months were comparable between the two groups. Regarding the aberrations, the results of Coma 90° in the SPT group were better than those in the non-SPT group (P ≤ 0.05), but the increase of RMS HOAs (root mean square higher order aberrations), Coma 0°, and spherical aberration postoperatively had no statistical difference between the two groups (P ≥ 0.05). Conclusions: Both aberration-free ASLA with and without SPT showed favorable safety, effectiveness, and predictability within 12 months for high myopia. And, ASLA using SPT might have potential advantages in the long-term visual quality.