S3NN: Time step reduction of spiking surrogate gradients for training energy efficient single-step spiking neural networks.

As the scales of neural networks increase, techniques that enable them to run with low computational cost and energy efficiency are required. From such demands, various efficient neural network paradigms, such as spiking neural networks (SNNs) or binary neural networks (BNNs), have been proposed. However, they have sticky drawbacks, such as degraded inference accuracy and latency. To solve these problems, we propose a single-step spiking neural network (S3NN), an energy-efficient neural network with low computational cost and high precision. The proposed S3NN processes the information between hidden layers by spikes as SNNs. Nevertheless, it has no temporal dimension so that there is no latency within training and inference phases as BNNs. Thus, the proposed S3NN has a lower computational cost than SNNs that require time-series processing. However, S3NN cannot adopt naïve backpropagation algorithms due to the non-differentiability nature of spikes. We deduce a suitable neuron model by reducing the surrogate gradient for multi-time step SNNs to a single-time step. We experimentally demonstrated that the obtained surrogate gradient allows S3NN to be trained appropriately. We also showed that the proposed S3NN could achieve comparable accuracy to full-precision networks while being highly energy-efficient.

Rethinking the Role of Normalization and Residual Blocks for Spiking Neural Networks.

Biologically inspired spiking neural networks (SNNs) are widely used to realize ultralow-power energy consumption. However, deep SNNs are not easy to train due to the excessive firing of spiking neurons in the hidden layers. To tackle this problem, we propose a novel but simple normalization technique called postsynaptic potential normalization. This normalization removes the subtraction term from the standard normalization and uses the second raw moment instead of the variance as the division term. The spike firing can be controlled, enabling the training to proceed appropriately, by conducting this simple normalization to the postsynaptic potential. The experimental results show that SNNs with our normalization outperformed other models using other normalizations. Furthermore, through the pre-activation residual blocks, the proposed model can train with more than 100 layers without other special techniques dedicated to SNNs.

A study on graphical model structure for representing statistical shape model of point distribution model.

In this article, the authors demonstrate that you can improve the performance of the registration of a point distribution model (PDM) by accurately estimating the structure of an undirected graphical model that represents the statistical shape model (SSM) of a target surface. Many existing methods for constructing SSMs determine the structure of the graphical model without analyzing the conditional dependencies among the points in PDM, though an edge in the PDM should link two nodes if and only if they are conditionally dependent. In this study, the authors employed four popular methods for estimating the structure of graphical model and obtained four different SSMs from an identical set of training surfaces. The registration performances of the SSMs were experimentally compared, and the results showed that the graphical lasso, which could estimate more accurate structure of the graphical model by avoiding the overfitting to the training data, outperformed the other methods.

Chronic total occlusion of the left main coronary artery with normal left ventricular motion: the occluded site confirmed by three-dimensional computed tomography.

Total occlusion of the left main trunk (LMT) frequently results in sudden cardiac death. As a result, it is rarely observed on coronary arteriogram. There are only a few reports on chronic total occlusion of the LMT. Most patients present with recent, severe angina, but it is not easy to distinguish chronic total occlusion of the LMT from other types of severe coronary heart diseases. Here, we report a very rare case of chronic total occlusion of the LMT. The patient is a 38-year-old female with a history of three normal deliveries. Chronic total occlusion of the LMT was suspected on coronary arteriogram 2 years previously in the other hospital; however, she continued working as a part-time employee at a supermarket. She was referred to our hospital because of slightly increased effort angina and shortness of breath. The final diagnosis and the site of occlusion were determined by three-dimensional computed tomography (3-D CT). The patient underwent coronary artery bypass graft (CABG) surgery, and ischemic symptoms completely disappeared.

Possible roles of angiotensin II-forming enzymes, angiotensin converting enzyme and chymase-like enzyme, in the human aneurysmal aorta.

Aortic aneurysm is a chronic degenerative condition associated with atherosclerosis. Recent studies have revealed that angiotensin (Ang) II plays important roles in atherosclerosis. In this study, to investigate the relationship between aortic aneurysm and Ang II, we measured the activities of the angiotensin (Ang) II-forming enzymes, angiotensin converting enzyme (ACE) and chymase-like enzyme, in human aneurysmal and control aortae. Aneurysmal aortic specimens were obtained from 16 aneurysm patients and control aortic specimens were obtained from 16 patients who underwent coronary artery bypass surgery (8 patients in each group were administered ACE inhibitors). The ACE and chymase-like enzyme activities were determined using extracts from vascular tissues. Both the ACE and chymase-like enzyme activities in the aneurysmal aortae were significantly higher than those in the control aortae (p < 0.01). In the patients treated with ACE inhibitors, the ACE activity in the aneurysmal aortae tended to be low, but the chymase-like enzyme activity tended to be high. In the aneurysmal aortae, the chymase-like enzyme activity in the adventitia was significantly higher than that in the intimal or medial layers (p < 0.01), while differences in ACE activity were not observed. Our results suggest that increases in local Ang II formation induced by chymase-like enzymes may play important roles in the pathogenesis of aneurysmal formation.

Increased local angiotensin II formation in aneurysmal aorta.

We investigated the levels and locations of angiotensin II-forming enzymes, angiotensin converting enzyme (ACE) and chymase, in aneurysmal and normal aortas. Aneurysmal aortic specimens (n = 14) were obtained at the time of operative aneurysm repair from 14 patients ranging in age from 57 to 84 y. Normal aortic specimens (n = 16) were obtained from 16 patients (48 to 72 y) who underwent coronary artery bypass surgery. The ACE and chymase activities were determined using each specimen. Sections of each specimen were immunostained with antibodies for ACE and chymase. The ACE activities in the aneurysmal and normal aortas were 0.82 +/- 0.10 and 0.14 +/- 0.05 mU/mg protein, respectively, and this difference was significant. The chymase activities in the aneurysmal and normal aortas were 17.9 +/- 2.40 and 1.02 +/- 0.18 mU/mg protein, respectively, and this difference was also significant. In the aneurysmal aorta, ACE-positive cells were detected with macrophages in the intima and media and chymase-positive cells were detected with mast cells in the media and adventitia, whereas positive ACE and chymase cells in the normal aorta were located only in the endothelium and adventitia, respectively. Angiotensin II-forming enzymes, chymase and ACE, were significantly increased in the aneurysmal aorta, and increased angiotensin II may be associated with the development of aneurysmal formations.