From the Cover: Convolutional networks for fast,energy-efficient neuromorphic computing

Deep networks are now able to achieve human-level performance on a broad spectrum of recognition tasks. Independently, neuromorphic computing has now demonstrated unprecedented energy-efficiency through a new chip architecture based on spiking neurons, low precision synapses, and a scalable communication network. Here, we demonstrate that neuromorphic computing, despite its novel architectural primitives, can implement deep convolution networks that (i) approach state-of-the-art classification accuracy across eight standard datasets encompassing vision and speech, (ii) perform inference while preserving the hardware’s underlying energy-efficiency and high throughput, running on the aforementioned datasets at between 1,200 and 2,600 frames/s and using between 25 and 275 mW (effectively >6,000 frames/s per Watt), and (iii) can be specified and trained using backpropagation with the same ease-of-use as contemporary deep learning. This approach allows the algorithmic power of deep learning to be merged with the efficiency of neuromorphic processors, bringing the promise of embedded, intelligent, brain-inspired computing one step closer.The human brain is capable of remarkable acts of perception while consuming very little energy. The dream of brain-inspired computing is to build machines that do the same, requiring high-accuracy algorithms and efficient hardware to run those algorithms. On the algorithm front, building on classic work on backpropagation (1), the neocognitron (2), and convolutional networks (3), deep learning has made great strides in achieving human-level performance on a wide range of recognition tasks (4). On the hardware front, building on foundational work on silicon neural systems (5), neuromorphic computing, using novel architectural primitives, has recently demonstrated hardware capable of running 1 million neurons and 256 million synapses for extremely low power (just 70 mW at real-time operation) (6). Bringing these approaches together holds the promise of a new generation of embedded, real-time systems, but first requires reconciling key differences in the structure and operation between contemporary algorithms and hardware. Here, we introduce and demonstrate an approach we call Eedn, energy-efficient deep neuromorphic networks, which creates convolutional networks whose connections, neurons, and weights have been adapted to run inference tasks on neuromorphic hardware.For structure, typical convolutional networks place no constraints on filter sizes, whereas neuromorphic systems can take advantage of blockwise connectivity that limits filter sizes, thereby saving energy because weights can now be stored in local on-chip memory within dedicated neural cores. Here, we present a convolutional network structure that naturally maps to the efficient connection primitives used in contemporary neuromorphic systems. We enforce this connectivity constraint by partitioning filters into multiple groups and yet maintain network integration by interspersing layers whose filter support region is able to cover incoming features from many groups by using a small topographic size (7).For operation, contemporary convolutional networks typically use high precision ( ≥ 32-bit) neurons and synapses to provide continuous derivatives and support small incremental changes to network state, both formally required for backpropagation-based gradient learning. In comparison, neuromorphic designs can use one-bit spikes to provide event-based computation and communication (consuming energy only when necessary) and can use low-precision synapses to colocate memory with computation (keeping data movement local and avoiding off-chip memory bottlenecks). Here, we demonstrate that by introducing two constraints into the learning rule—binary-valued neurons with approximate derivatives and trinary-valued ({−1,0,1}) synapses—it is possible to adapt backpropagation to create networks directly implementable using energy efficient neuromorphic dynamics. This approach draws inspiration from the spiking neurons and low-precision synapses of the brain (8) and builds on work showing that deep learning can create networks with constrained connectivity (9), low-precision synapses (10, 11), low-precision neurons (12–14), or both low-precision synapses and neurons (15, 16). For input data, we use a first layer to transform multivalued, multichannel input into binary channels using convolution filters that are learned via backpropagation (12, 16) and whose output can be sent on chip in the form of spikes. These binary channels, intuitively akin to independent components (17) learned with supervision, provide a parallel distributed representation to carry out high-fidelity computation without the need for high-precision representation.Critically, we demonstrate that bringing the above innovations together allows us to create networks that approach state-of-the-art accuracy performing inference on eight standard datasets, running on a neuromorphic chip at between 1,200 and 2,600 frames/s (FPS), using between 25 and 275 mW. We further explore how our approach scales by simulating multichip configurations. Ease-of-use is achieved using training tools built from existing, optimized deep learning frameworks (18), with learned parameters mapped to hardware using a high-level deployment language (19). Although we choose the IBM TrueNorth chip (6) for our example deployment platform, the essence of our constructions can apply to other emerging neuromorphic approaches (20–23) and may lead to new architectures that incorporate deep learning and efficient hardware primitives from the ground up.     

Umbilical cord blood transplantation--how, when and for whom?

In recent years, umbilical cord blood (UCB) has emerged as a feasible alternative source of hematopoietic progenitors (CD34+) for allogeneic stem cell transplantation, mainly in patients who lack HLA-matched marrow donors. Since the first case reported in 1998, more than 3500 patients have received UCB transplants for a variety of malignant and non-malignant diseases. The vast majority of recipients were children with an average weight of 20 kg; however, more than 500 UCB transplantations (UCBTs) have already been performed in adults. The "naive" nature of UCB lymphocytes also permits the use of HLA-mismatched grafts at 1-2 loci without higher risk for severe graft versus host disease (GvHD) relative to bone marrow transplantation (BMT) from a full matched unrelated donor. Furthermore, UCB is rich in primitive CD16(-)CD56++ NK cells, which possess impressive proliferative and cytotoxic capacities and can be induced to expand using IL-12 or IL-15, so as to mount a substantial graft versus leukemia (GvL) effect. The main disadvantage of UCB is the low stem cell yields, resulting in higher rates of graft failure as well as delayed time to engraftment compared to BMT. One rational approach to overcome this limitation involves ex vivo expansion of UCB derived hematopoietic precursors. In this review we tried to answer the question: UCBT how, when and for whom. This procedure is mostly applicable for children and especially those with indication for full allogeneic transplantation but who lack a matched sibling donor. Experimental approaches including ex vivo expansion of CB with cocktail of hematopoietic growth factors, with or without differentiation blocking agents, co-transplantation of haploidentical and CB cells or co-transfusion of CB and mesenchymal cells may enable successful UCBT in adults and probably will result in expanding the indication to solid tumors or autoimmune disorders.     

Varicella zoster infection and pulmonary complications

A 34-year-old immunocompetent man with varicella zoster (VZ) infection developed deep vein thrombosis and pulmonary embolism after suffering severe pneumonitis. He recovered after treatment with acyclovir, high-dose steroids, and ventilatory support. The endothelial damage could be a direct link between VZ pneumonitis and pulmonary emboli.     

Does the level of amniotic fluid have an effect on the accuracy of sonographic estimated fetal weight at term?

Objective: Controversy exists concerning the impact of amniotic fluid index (AFI) on the accuracy of sonographic estimation of fetal weight (EFW). Thus, we aimed to evaluate whether differences in AFI has an influence on the accuracy of sonographic EFW.

Methods: All term, singleton pregnancies which underwent a sonographic EFW and measurement of AFI within a week from delivery were included. Cases were stratified into three categories according to AFI: (1) Normal AFI (51–249?mm), (2) Oligohydramnios (AFI?≤?50?mm) and (3) Polyhydramnios (AFI?≥?250?mm). Inaccurate EFW was defined if there was more than 15% difference between sonographic EFW and actual birthweight.

Results: Overall, 1746 pregnancies were identified (1096 with normal AFI, 455 with oligohydramnios and 195 with polyhydramnios). Mean AFI was 115.8?±?60?mm, 28.1?±?13?mm and 293?±?35?mm, p?<?0.001, and mean sonographic EFW was 3182.5?±?573?g, 3118.8?±?517?g and 3713.2?±?461?g, p?<?0.001, respectively. Demographic data and gestational age at delivery were similar. Mean birthweight was 3221.7?±?535?g, 3132.5?±?505?g and 3654.1?±?480?g, p?<?0.001, respectively. The rate of inaccurate EFW was similar between the groups (8.4%, 8.7% and 9.7%, p?=?0.19, respectively). On multivariate analysis, AFI was not associated with EFW inaccuracy (OR 1.01, 95% C.I 0.67–1.54, p?=?0.93).

Conclusion: AFI has limited impact on the percentage of errors in sonographic fetal weight estimation a week prior delivery.     

Assessment of the coagulation profile in hemato-oncological patients receiving ATG-based conditioning treatment for allogeneic stem cell transplantation

Antithymocyte globulin (ATG) is increasingly used in pre-allogeneic stem cell transplantation (allo-SCT) conditioning regimens to prevent graft rejection and graft-versus-host disease. However, ATG was also found to be associated with increased incidence of thrombosis during organ transplantation. In the present study, we tested the coagulation status of 21 patients with hematologic malignancies undergoing allo-SCT who received ATG-based (11 patients) or non-ATG-based (10) conditioning treatment. We assessed several thrombophilia markers as well as circulating total and endothelial microparticles (TMP/EMP) and soluble CD40 ligand (CD40L). No significant difference in the mean values of prothrombin time, partial thromboplastin time, fibrinogen, antithrombin, protein C, protein S, thrombin-antithrombin III complex, homocysteine levels, prevalence of genetic thrombophilia markers and levels of EMP, TMP or CD40L was observed between the ATG-treated and ATG-untreated patients, as well as before and after conditioning in each group separately. Platelet counts decreased significantly in ATG-treated patients; however, this decrease was not associated with clinical or laboratory evidence of disseminated intravascular coagulation. No patient developed thromboembolic event or veno-occlusive liver disease. Our results suggest that allo-SCT is not associated with increased hypercoagulability and addition of ATG to conditioning regimen has no significant procoagulant effect.     

Emergency pediatric radiology imaging trends for non-COVID-19-related illnesses through different stages of the pandemic

Emergency Radiology - To evaluate how the COVID-19 pandemic affected the imaging utilization patterns for non-COVID-19-related illness in a pediatric emergency department (ED). We retrospectively...     

Effects of lovastatin therapy on LDL receptor activity in circulating monocytes and on structure and composition of plasma lipoproteins

The effect of lovastatin therapy on LDL-receptor activity in fresh monocytes and on the structure and composition of lipoproteins was determined in 9 patients with familial hypercholesterolemia (FH) and 8 patients with non-familial hypercholesterolemia (NFH). Lovastatin reduced LDL-cholesterol levels by 34.8 and 47.5%, respectively, in the 2 groups of patients, and plasma apo B levels by 33.3 and 42.5%. LDL receptor activity in fresh monocytes increased by 53% and 86% respectively. HDL-cholesterol and plasma apo A-I levels increased only in the NFH group, by 10.2 and 7.1%. Lipoproteins were separated by centrifugation on a zonal rotor. Except for the intermediate density lipoprotein (IDL) fraction, no changes were observed in the structure and composition of the various lipoproteins. The investigations thus demonstrated that lovastatin therapy is associated with a measurable and significant increase of LDL-receptor activity in circulating monocytes that may contribute to the lipid lowering action of the drug.     

Chemiluminescence response of polymorphonuclear leucocytes to serum-opsonized zymosan--a predictive parameter for infections and prognosis in multiple myeloma patients

Serum opsonization capability and phagocytosis by polymorphonuclear leucocytes (PMNL) were investigated in 37 patients with multiple myeloma (MM), 32 patients with chronic lymphocytic leukaemia (CLL), and 30 healthy controls. Opsonophagocytosis was assessed by measuring the chemiluminescence (CL) response of controls' and patients' PMNL to control- and patient-serum-opsonized zymosan particles. Control PMNL CL responses were significantly reduced both by MM and CLL sera. Normal sera restored the CL responses in MM patients, but only partially corrected the CL responses in CLL patients. Significant correlations were found between the CL responses in MM patients and four clinical parameters: clinical stage, presence of disease-related symptoms, incidence of bacterial infections, and outcome. The findings suggest a defect in MM and CLL sera opsonization ability which, at least in part, could account for the increased susceptibility to infections observed in these patients. Furthermore, the assessment of the CL response in MM patients is recommended as a simple and useful clinical tool for prediction of susceptibility to infection, and prognosis.