Machine intelligence has become a driving factor in modern society. However, its demand outpaces the underlying electronic technology due to limitations given by fundamental physics, such as capacitive charging of wires, but also by system architecture of storing and handling data, both driving recent trends toward processor heterogeneity. Task-specific accelerators based on free-space optics bear fundamental homomorphism for massively parallel and real-time information processing given the wave nature of light. However, initial results are frustrated by data handling challenges and slow optical programmability. Here we introduce a novel amplitude-only Fourier-optical processor paradigm capable of processing large-scale ∼ ( 1000 × 1000 ) matrices in a single time step and 100 µs-short latency. Conceptually, the information flow direction is orthogonal to the two-dimensional programmable network, which leverages 10 6 parallel channels of display technology, and enables a prototype demonstration performing convolutions as pixelwise multiplications in the Fourier domain reaching peta operations per second throughputs. The required real-to-Fourier domain transformations are performed passively by optical lenses at zero-static power. We exemplary realize a convolutional neural network (CNN) performing classification tasks on 2 megapixel large matrices at 10 kHz rates, which latency-outperforms current graphic processing unit and phase-based display technology by 1 and 2 orders of magnitude, respectively. Training this optical convolutional layer on image classification tasks and utilizing it in a hybrid optical-electronic CNN, shows classification accuracy of 98% (Modified National Institute of Standards and Technology) and 54% (CIFAR-10). Interestingly, the amplitude-only CNN is inherently robust against coherence noise in contrast to phase-based paradigms and features a delay over 2 orders of magnitude lower than liquid-crystal-based systems. Such an amplitude-only massively parallel optical compute paradigm shows that the lack of phase information can be accounted for via training, thus opening opportunities for high-throughput accelerator technology for machine intelligence with applications in network-edge processing, in data centers, or in pre-processing information or filtering toward near-real-time decision making.
The performance of motor imagery (MI) based Brain-computer interfacing (BCI) is easily affected by noise and redundant information that exists in the multi-channel electroencephalogram (EEG). To solve this problem, many temporal and spatial feature based channel selection methods have been proposed. However, temporal and spatial features do not accurately reflect changes in the power of the oscillatory EEG. Thus, spectral features of MI-related EEG signals may be useful for channel selection. Bispectrum analysis is a technique developed for extracting non-linear and non-Gaussian information from non-linear and non-Gaussian signals. The features extracted from bispectrum analysis can provide frequency domain information about the EEG. Therefore, in this study, we propose a bispectrumbased channel selection (BCS) method for MI-based BCI. The proposed method uses the sum of logarithmic amplitudes (SLA) and the first order spectral moment (FOSM) features extracted from bispectrum analysis to select EEG channels without redundant information. Three public BCI competition datasets (BCI competition IV dataset 1, BCI competition III dataset IVa and BCI competition III dataset IIIa) were used to validate the effectiveness of our proposed method. The results indicate that our BCS method outperforms use of all channels (83.8% vs 69.4%, 86.3% vs 82.9% and 77.8% vs 68.2%, respectively). Furthermore, compared to the other state-of-the-art methods, our BCS method also
P300-based brain-computer interfaces (BCIs) provide an additional communication channel for individuals with communication disabilities. In general, P300-based BCIs need to be trained, offline, for a considerable period of time, which causes users to become fatigued. This reduces the efficiency and performance of the system. In order to shorten calibration time and improve system performance, we introduce the concept of a generic model set. We used ERP data from 116 participants to train the generic model set. The resulting set consists of ten models, which are trained by weighted linear discriminant analysis (WLDA). Twelve new participants were then invited to test the validity of the generic model set. The results demonstrated that all new participants matched the best generic model. The resulting mean classification accuracy equaled 80% after online training, an accuracy that was broadly equivalent to the typical training model method. Moreover, the calibration time was shortened by 70.7% of the calibration time of the typical model method. In other words, the best matching model method only took 81s to calibrate, while the typical model method took 276s. There were also significant differences in both accuracy and raw bit rate between the best and the worst matching model methods. We conclude that the strategy of combining the generic models with online training is easily accepted and achieves higher levels of user satisfaction (as measured by subjective reports). Thus, we provide a valuable new strategy for improving the performance of P300-based BCI.
The motor imagery (MI) based brain-computer interfaces (BCIs) have been proposed as a potential physical rehabilitation technology. However, the low classification accuracy achievable with MI tasks is still a challenge when building effective BCI systems. We propose a novel MI classification model based on measurement of functional connectivity between brain regions and graph theory. Specifically, motifs describing local network structures in the brain are extracted from functional connectivity graphs. A graph embedding model called Ego-CNNs is then used to build a classifier, which can convert the graph from a structural representation to a fixed-dimensional vector for detecting critical structure in the graph.We validate our proposed method on four datasets, and the results show that our proposed method produces high classification accuracies in two-class classification tasks (92.8% for dataset 1, 93.4% for dataset 2, 96.5% for dataset 3, and 80.2% for dataset 4) and multiclass classification tasks (90.33% for dataset 1). Our proposed method achieves a mean Kappa value of 0.88 across nine participants, which is superior to other methods we compared it to. These results indicate that there is a local structural difference in functional connectivity graphs extracted under different motor imagery tasks. Our proposed method has great potential for motor imagery classification in future studies.
Decision-making through artificial neural networks with minimal latency is critical for numerous applications such as navigation, tracking, and real-time machine action systems. This requires machine learning hardware to process multidimensional data at high throughput. Unfortunately, handling convolution operations, the primary computational tool for data classification tasks, obeys challenging runtime complexity scaling laws. However, homomorphically implementing the convolution theorem in a Fourier optics display light processor can achieve a non-iterative (1) runtime complexity for data inputs beyond 1,000 × 1,000 large matrices. Following this approach, here we demonstrate data streaming multi-kernel image batching using a Fourier Convolutional Neural Network (FCNN) accelerator. We show image batch processing of large-scale matrices as 2 million dot product multiplications performed by a digital light processing module in the Fourier domain. Furthermore, we further parallelize this optical FCNN system by exploiting multiple spatially parallel diffraction orders, achieving a 98x throughput improvement over state-of-the-art FCNN accelerators. A comprehensive discussion of the practical challenges associated with working at the edge of system capabilities highlights the problem of crosstalk and resolution scaling laws in the Fourier domain. Accelerating convolution by exploiting massive parallelism in display technology brings non-Van Neumann-based machine learning acceleration.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.