with memory
Recently Published Documents





2022 ◽  
Vol 19 (1) ◽  
pp. 1-23
Yaosheng Fu ◽  
Evgeny Bolotin ◽  
Niladrish Chatterjee ◽  
David Nellans ◽  
Stephen W. Keckler

As GPUs scale their low-precision matrix math throughput to boost deep learning (DL) performance, they upset the balance between math throughput and memory system capabilities. We demonstrate that a converged GPU design trying to address diverging architectural requirements between FP32 (or larger)-based HPC and FP16 (or smaller)-based DL workloads results in sub-optimal configurations for either of the application domains. We argue that a C omposable O n- PA ckage GPU (COPA-GPU) architecture to provide domain-specialized GPU products is the most practical solution to these diverging requirements. A COPA-GPU leverages multi-chip-module disaggregation to support maximal design reuse, along with memory system specialization per application domain. We show how a COPA-GPU enables DL-specialized products by modular augmentation of the baseline GPU architecture with up to 4× higher off-die bandwidth, 32× larger on-package cache, and 2.3× higher DRAM bandwidth and capacity, while conveniently supporting scaled-down HPC-oriented designs. This work explores the microarchitectural design necessary to enable composable GPUs and evaluates the benefits composability can provide to HPC, DL training, and DL inference. We show that when compared to a converged GPU design, a DL-optimized COPA-GPU featuring a combination of 16× larger cache capacity and 1.6× higher DRAM bandwidth scales per-GPU training and inference performance by 31% and 35%, respectively, and reduces the number of GPU instances by 50% in scale-out training scenarios.

2022 ◽  
Vol 119 (3) ◽  
pp. e2107661119
William P. Dempsey ◽  
Zhuowei Du ◽  
Anna Nadtochiy ◽  
Colton D. Smith ◽  
Karl Czajkowski ◽  

Defining the structural and functional changes in the nervous system underlying learning and memory represents a major challenge for modern neuroscience. Although changes in neuronal activity following memory formation have been studied [B. F. Grewe et al., Nature 543, 670–675 (2017); M. T. Rogan, U. V. Stäubli, J. E. LeDoux, Nature 390, 604–607 (1997)], the underlying structural changes at the synapse level remain poorly understood. Here, we capture synaptic changes in the midlarval zebrafish brain that occur during associative memory formation by imaging excitatory synapses labeled with recombinant probes using selective plane illumination microscopy. Imaging the same subjects before and after classical conditioning at single-synapse resolution provides an unbiased mapping of synaptic changes accompanying memory formation. In control animals and animals that failed to learn the task, there were no significant changes in the spatial patterns of synapses in the pallium, which contains the equivalent of the mammalian amygdala and is essential for associative learning in teleost fish [M. Portavella, J. P. Vargas, B. Torres, C. Salas, Brain Res. Bull. 57, 397–399 (2002)]. In zebrafish that formed memories, we saw a dramatic increase in the number of synapses in the ventrolateral pallium, which contains neurons active during memory formation and retrieval. Concurrently, synapse loss predominated in the dorsomedial pallium. Surprisingly, we did not observe significant changes in the intensity of synaptic labeling, a proxy for synaptic strength, with memory formation in any region of the pallium. Our results suggest that memory formation due to classical conditioning is associated with reciprocal changes in synapse numbers in the pallium.

Natasha K. Brusco ◽  
Helen Kugler ◽  
Fiona Dufler ◽  
Annemarie L. Lee ◽  
Brianna Walpole ◽  

Objective: To test the feasibility, safety and effectiveness of the My Therapy programme for inpatients with mild-moderate cognitive impairment. Design: Observational pilot study. Patients: Rehabilitation inpatients with mild-moderate cognitive impairment. Methods: During their inpatient admission, participants received My Therapy, a programme that can increase the dose of rehabilitation through independent self-practice of exercises, outside of supervised therapy. Outcomes included My Therapy participation, falls, Functional Independence Measure (FIM) and 10-m walk test. Outcomes were compared with those of participants without cognitive impairment from the original My Therapy study (n = 116) using χ2 and independent t-tests.  Results: Eight participants with mild-moderate cognitive impairment (mean (standard deviation (SD)) age 89.6 years (4.8); 3 women) were included. All participants completed the My Therapy programme on at least one day of their admission, with no associated falls. Participants had an 8.4 s (SD 5.1) reduction in their 10-m walk test and a 21.5 point (SD 11.1) improvement on FIM scores from admission to discharge. There were no significant between-group differences in feasibility, safety or effectiveness for participants with and without cognitive impairment. Conclusion: This pilot study has shown that including exercise self-management as part of inpatient rehabilitation is feasible, safe and effective for patients with cognitive impairment.    Lay Abstract This study aimed to determine whether it was practical, safe and effective for patients in a rehabilitation hospital with memory or thinking problems to participate in a programme called My Therapy. My Therapy aimed to increase the dose of rehabilitation through independent self-practice of exercises, outside of supervised therapy sessions. There were 8 participants in the study and all of them reported completing the My Therapy programme on at least one day of their rehabilitation stay. There were no falls relating to My Therapy participation. Participants improved their walking speed and function during their rehabilitation stay. There were no differences in the results between people with and without memory or thinking problems, in terms of practicality, safety or effectiveness. This study has shown that including exercise self-management as part of rehabilitation is practical, safe and effective for patients with memory or thinking problems. 

Neurology ◽  
2022 ◽  
pp. 10.1212/WNL.0000000000013299
Pontus Tideman ◽  
Erik Stomrud ◽  
Antoine Leuzy ◽  
Niklas Mattsson-Carlgren ◽  
Sebastian Palmqvist ◽  

Background and Objectives:The neuropathological changes underlying Alzheimer´s disease (AD) start before overt cognitive symptoms arise, but it is not well-known how they relate to the first subtle cognitive changes. The objective for this study was to examine the independent associations of the AD hallmarks β-amyloid (Aβ), tau, and neurodegeneration with different cognitive domains in cognitively unimpaired (CU) individuals.Methods:In this cross-sectional study, CU participants from the prospective BioFINDER-2 study were included. All had CSF biomarkers (Aβ42 and P-tau181), MRI (cortical thickness of AD-susceptible regions), Aβ-PET (neocortical uptake), tau-PET (entorhinal uptake), and cognitive test data for i) memory, ii) executive function, iii) verbal function, iv), and visuospatial function. Multivariable linear regression models were performed, using either CSF Aβ42, P-tau181 and cortical thickness or Aβ-PET, tau-PET, and cortical thickness, as predictors of cognitive function. The results were validated in an independent cohort (ADNI).Results:316 CU participants were included from the BioFINDER-2 study. Abnormal Aβ-status was independently associated with the executive measure, regardless of modality (CSF Aβ42 β=0.128, p=0.024; Aβ-PET β=0.124, p=0.049), while tau was independently associated with memory (CSF P-tau181 β=0.132, p=0.018; tau-PET β=0.189, p=0.002). Cortical thickness was independently associated with the executive measure and verbal fluency in both models (p=0.005-0.018). To examine the relationships in the earliest stage of preclinical AD, only participants with normal biomarkers of tau and neurodegeneration were included (n=217 CSF-based; n=246 PET-based). Again, Aβ-status was associated with executive function (CSF Aβ42, β=0.189, p=0.005; Aβ-PET, β=0.146, p=0.023), but not with other cognitive domains. The results were overall replicated in the ADNI cohort (n=361).Discussion:These findings suggest that Aβ is independently associated with worse performance on an executive measure but not with memory performance, which instead is associated with tau pathology. This may have implications for early preclinical AD screening and outcome measures in AD trials targeting Aβ pathology.

Mathematics ◽  
2022 ◽  
Vol 10 (2) ◽  
pp. 224
Paola Rubbioni

In this paper, we study a semilinear integro-differential inclusion in Banach spaces, under the action of infinitely many impulses. We provide the existence of mild solutions on a half-line by means of the so-called extension-with-memory technique, which consists of breaking down the problem in an iterate sequence of non-impulsive Cauchy problems, each of them originated by a solution of the previous one. The key that allows us to employ this method is the definition of suitable auxiliary set-valued functions that imitate the original set-valued nonlinearity at any step of the problem’s iteration. As an example of application, we deduce the controllability of a population dynamics process with distributed delay and impulses. That is, we ensure the existence of a pair trajectory-control, meaning a possible evolution of a population and of a feedback control for a system that undergoes sudden changes caused by external forces and depends on its past with fading memory.

2022 ◽  
Ondrej Kucera ◽  
Jeremie Gaillard ◽  
Christophe Guerin ◽  
Manuel Thery ◽  
Laurent Blanchoin

Active cytoskeletal materials in vitro demonstrate self-organising properties similar to those observed in their counterparts in cells. However, the search to emulate phenomena observed in the living matter has fallen short of producing a cytoskeletal network that would be structurally stable yet possessing adaptive plasticity. Here, we address this challenge by combining cytoskeletal polymers in a composite, where self-assembling microtubules and actin filaments collectively self-organise due to the activity of microtubules-percolating molecular motors. We demonstrate that microtubules spatially organise actin filaments that in turn guide microtubules. The two networks align in an ordered fashion using this feedback loop. In this composite, actin filaments can act as structural memory and, depending on the concentration of the components, microtubules either write this memory or get guided by it. The system is sensitive to external stimuli suggesting possible autoregulatory behaviour in changing mechanochemical environment. We thus establish artificial active actin-microtubule composite as a system demonstrating architectural stability and plasticity.

2022 ◽  
Vol 15 ◽  
Vivek Parmar ◽  
Bogdan Penkovsky ◽  
Damien Querlioz ◽  
Manan Suri

With recent advances in the field of artificial intelligence (AI) such as binarized neural networks (BNNs), a wide variety of vision applications with energy-optimized implementations have become possible at the edge. Such networks have the first layer implemented with high precision, which poses a challenge in deploying a uniform hardware mapping for the network implementation. Stochastic computing can allow conversion of such high-precision computations to a sequence of binarized operations while maintaining equivalent accuracy. In this work, we propose a fully binarized hardware-friendly computation engine based on stochastic computing as a proof of concept for vision applications involving multi-channel inputs. Stochastic sampling is performed by sampling from a non-uniform (normal) distribution based on analog hardware sources. We first validate the benefits of the proposed pipeline on the CIFAR-10 dataset. To further demonstrate its application for real-world scenarios, we present a case-study of microscopy image diagnostics for pathogen detection. We then evaluate benefits of implementing such a pipeline using OxRAM-based circuits for stochastic sampling as well as in-memory computing-based binarized multiplication. The proposed implementation is about 1,000 times more energy efficient compared to conventional floating-precision-based digital implementations, with memory savings of a factor of 45.

Sign in / Sign up

Export Citation Format

Share Document