Spiking neural networks (SNNs) incorporating biologically plausible neurons hold great promise because of their unique temporal dynamics and energy efficiency. However, SNNs have developed separately ...from artificial neural networks (ANNs), limiting the impact of deep learning advances for SNNs. Here, we present an alternative perspective of the spiking neuron that incorporates its neural dynamics into a recurrent ANN unit called a spiking neural unit (SNU). SNUs may operate as SNNs, using a step function activation, or as ANNs, using continuous activations. We demonstrate the advantages of SNU dynamics through simulations on multiple tasks and obtain accuracies comparable to, or better than, those of ANNs. The SNU concept enables an efficient implementation with in-memory acceleration for both training and inference. We experimentally demonstrate its efficacy for a music-prediction task in an in-memory-based SNN accelerator prototype using 52,800 phase-change memory devices. Our results open up an avenue for broad adoption of biologically inspired neural dynamics in challenging applications and acceleration with neuromorphic hardware.Spiking neural networks and in-memory computing are both promising routes towards energy-efficient hardware for deep learning. Woźniak et al. incorporate the biologically inspired dynamics of spiking neurons into conventional recurrent neural network units and in-memory computing, and show how this allows for accurate and energy-efficient deep learning.
Plasticity circuits in the brain are known to be influenced by the distribution of the synaptic weights through the mechanisms of synaptic integration and local regulation of synaptic strength. ...However, the complex interplay of stimulation-dependent plasticity with local learning signals is disregarded by most of the artificial neural network training algorithms devised so far. Here, we propose a novel biologically inspired optimizer for artificial and spiking neural networks that incorporates key principles of synaptic plasticity observed in cortical dendrites: GRAPES (Group Responsibility for Adjusting the Propagation of Error Signals). GRAPES implements a weight-distribution-dependent modulation of the error signal at each node of the network. We show that this biologically inspired mechanism leads to a substantial improvement of the performance of artificial and spiking networks with feedforward, convolutional, and recurrent architectures, it mitigates catastrophic forgetting, and it is optimally suited for dedicated hardware implementations. Overall, our work indicates that reconciling neurophysiology insights with machine intelligence is key to boosting the performance of neural networks.
Abstract
Visual oddity task was conceived to study universal ethnic-independent analytic intelligence of humans from a perspective of comprehension of spatial concepts. Advancements in artificial ...intelligence led to important breakthroughs, yet excelling at such abstract tasks remains challenging. Current approaches typically resort to non-biologically-plausible architectures with ever-growing models consuming substantially more energy than the brain. Motivated by the brain’s efficiency and reasoning capabilities, we present a biologically inspired system that receives inputs from synthetic eye movements – reminiscent of saccades, and processes them with neuronal units incorporating dynamics of neocortical neurons. We introduce a procedurally generated visual oddity dataset to train an architecture extending conventional relational networks and our proposed system. We demonstrate that both approaches are capable of abstract problem-solving at high accuracy, and we uncover that both share the same essential underlying mechanism of reasoning in seemingly unrelated aspects of their architectures. Finally, we show that the biologically inspired network achieves superior accuracy, learns faster and requires fewer parameters than the conventional network.
A novel scan trajectory for high-speed scanning probe microscopy is presented in which the probe follows a two-dimensional Lissajous pattern. The Lissajous pattern is generated by actuating the ...scanner with two single-tone harmonic waveforms of constant frequency and amplitude. Owing to the extremely narrow frequency spectrum, high imaging speeds can be achieved without exciting the unwanted resonant modes of the scanner and without increasing the sensitivity of the feedback loop to the measurement noise. The trajectory also enables rapid multiresolution imaging, providing a preview of the scanned area in a fraction of the overall scan time. We present a procedure for tuning the spatial and the temporal resolution of Lissajous trajectories and show experimental results obtained on a custom-built atomic force microscope (AFM). Real-time AFM imaging with a frame rate of 1 frame s−1 is demonstrated.
At the algorithmic level, neuro-inspired learning paradigms are based on the insight that the brain continuously processes incoming information and is able to adapt to changing conditions. ...online ...learning, learning-to-learn, and unsupervised learning provide the main conceptual platforms for the design of low-power, accurate and reliable neuromorphic computing systems. In recent years, ANN-to-SNN conversion techniques have enabled the design of SNNs, starting from well-known ANN architectures, that offer lower computation cost compared to their non-spiking counterparts. ...the concept of in-memory computing, which aims at co-locating the memory and processing units, has recently demonstrated that substantial acceleration may be achieved for ANNs. ...Datta et al. propose a deep SNN for 3D image recognition using algorithmic and hardware co-design approaches, namely quantization-aware backpropagation and processing-in-memory (PIM) architecture.
Oncogenic gene fusions drive many human cancers, but tools to more quickly unravel their functional contributions are needed. Here we describe methodology permitting fusion gene construction for ...functional evaluation. Using this strategy, we engineered the known fusion oncogenes,
, and
as well as 20 previously uncharacterized fusion genes identified in The Cancer Genome Atlas datasets. In addition to confirming oncogenic activity of the known fusion oncogenes engineered by our construction strategy, we validated five novel fusion genes involving
, and
kinases that exhibited potent transforming activity and conferred sensitivity to FDA-approved kinase inhibitors. Our fusion construction strategy also enabled domain-function studies of
fusion genes. Our results confirmed other reports that the transforming activity of
fusions results from truncation-mediated loss of inhibitory domains within the N-terminus of the BRAF protein.
mutations residing within this inhibitory region may provide a means for BRAF activation in cancer, therefore we leveraged the modular design of our fusion gene construction methodology to screen N-terminal domain mutations discovered in tumors that are wild-type at the
mutation hotspot, V600. We identified an oncogenic mutation, F247L, whose expression robustly activated the MAPK pathway and sensitized cells to BRAF and MEK inhibitors. When applied broadly, these tools will facilitate rapid fusion gene construction for subsequent functional characterization and translation into personalized treatment strategies.
.
This paper presents a dual-stage approach to nanopositioning in which the tradeoff between the scanner speed and range is addressed by combining a slow, large-range scanner with a short-range scanner ...optimized for high-speed, high-resolution positioning. We present the design, finite-element simulations, and experimental characterization of a fast custom-built short-range scanner. The short-range scanner is based on electromagnetic actuation to provide high linearity, has a clean, high-bandwidth dynamical response and is equipped with a low-noise magnetoresistance-based sensor. By using advanced noise-resilient feedback controllers, the dual-stage system allows large-range positioning with subnanometer closed-loop resolution over a wide bandwidth. Experimental results are presented in which the dual-stage scanner system is used for imaging in a custom-built atomic force microscope.
Optical identification is often done with spatial or temporal visual pattern recognition and localization. Temporal pattern recognition, depending on the technology, involves a trade-off between ...communication frequency, range, and accurate tracking. We propose a solution with light-emitting beacons that improves this trade-off by exploiting fast event-based cameras and, for tracking, sparse neuromorphic optical flow computed with spiking neurons. The system is embedded in a simulated drone and evaluated in an asset monitoring use case. It is robust to relative movements and enables simultaneous communication with, and tracking of, multiple moving beacons. Finally, in a hardware lab prototype, we demonstrate for the first time beacon tracking performed simultaneously with state-of-the-art frequency communication in the kHz range.
Genomic rearrangements exert a heavy influence on the molecular landscape of cancer. New analytical approaches integrating somatic structural variants (SSVs) with altered gene features represent a ...framework by which we can assign global significance to a core set of genes, analogous to established methods that identify genes non-randomly targeted by somatic mutation or copy number alteration. While recent studies have defined broad patterns of association involving gene transcription and nearby SSV breakpoints, global alterations in DNA methylation in the context of SSVs remain largely unexplored.
By data integration of whole genome sequencing, RNA sequencing, and DNA methylation arrays from more than 1400 human cancers, we identify hundreds of genes and associated CpG islands (CGIs) for which the nearby presence of a somatic structural variant (SSV) breakpoint is recurrently associated with altered expression or DNA methylation, respectively, independently of copy number alterations. CGIs with SSV-associated increased methylation are predominantly promoter-associated, while CGIs with SSV-associated decreased methylation are enriched for gene body CGIs. Rearrangement of genomic regions normally having higher or lower methylation is often involved in SSV-associated CGI methylation alterations. Across cancers, the overall structural variation burden is associated with a global decrease in methylation, increased expression in methyltransferase genes and DNA damage response genes, and decreased immune cell infiltration.
Genomic rearrangement appears to have a major role in shaping the cancer DNA methylome, to be considered alongside commonly accepted mechanisms including histone modifications and disruption of DNA methyltransferases.
PDF
