Spiking neural networks (SNNs) are positioned to enable spatio-temporal information processing and ultra-low power event-driven neuromorphic hardware. However, SNNs are yet to reach the same performances of conventional deep artificial neural networks (ANNs), a long-standing challenge due to complex dynamics and non-differentiable spike events encountered in training. The existing SNN error backpropagation (BP) methods are limited in terms of scalability, lack of proper handling of spiking discontinuities, and/or mismatch between the rate coded loss function and computed gradient. We present a hybrid macro/micro level backpropagation (HM2-BP) algorithm for training multi-layer SNNs. The temporal effects are precisely captured by the proposed spike-train level post-synaptic potential (S-PSP) at the microscopic level. The rate-coded errors are defined at the macroscopic level, computed and back-propagated across both macroscopic and microscopic levels. Different from existing BP methods, HM2-BP directly computes the gradient of the rate-coded loss function w.r.t tunable parameters. We evaluate the proposed HM2-BP algorithm by training deep fully connected and convolutional SNNs based on the static MNIST [14] and dynamic neuromorphic N-MNIST [26]. HM2-BP achieves an accuracy level of 99:49% and 98:88% for MNIST and N-MNIST, respectively, outperforming the best reported performances obtained from the existing SNN BP algorithms. Furthermore, the HM2-BP produces the highest accuracies based on SNNs for the EMNIST [3] dataset, and leads to high recognition accuracy for the 16-speaker spoken English letters of TI46 Corpus [16], a challenging spatio-temporal speech recognition benchmark for which no prior success based on SNNs was reported. It also achieves competitive performances surpassing those of conventional deep learning models when dealing with asynchronous spiking streams.
more »
« less
Conflict Forecasting with Event Data and Spatio-Temporal Graph Convolutional Networks
This paper explores three different model components to improve predictive performance over the ViEWS benchmark: a class of neural networks that account for spatial and temporal dependencies; the use of CAMEO-coded event data; and the continuous rank probability score (CRPS), which is a proper scoring metric. We forecast changes in state based violence across Africa at the grid-month level. The results show that spatio-temporal graph convolutional neural network models offer consistent improvements over the benchmark. The CAMEO-coded event data sometimes improve performance, but sometimes decrease performance. Finally, the choice of performance metric, whether it be the mean squared error or a proper metric such as the CRPS, has an impact on model selection. Each of these components–algorithms, measures, and metrics–can improve our forecasts and understanding of violence.
more »
« less
- Award ID(s):
- 1931541
- NSF-PAR ID:
- 10376284
- Date Published:
- Journal Name:
- International Interactions
- ISSN:
- 0305-0629
- Page Range / eLocation ID:
- 1 to 23
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
Abstract Numerical weather prediction models and high-performance computing have significantly improved our ability to model near-surface variables, but their uncertainty quantification still remains a challenging task. Ensembles are usually produced to depict a series of possible future states of the atmosphere, as a means to quantify the prediction uncertainty, but this requires multiple instantiation of the model, leading to an increased computational cost. Weather analogs, alternatively, can be used to generate ensembles without repeated model runs. The analog ensemble (AnEn) is a technique to identify similar weather patterns for near-surface variables and quantify forecast uncertainty. Analogs are chosen based on a similarity metric that calculates the weighted multivariate Euclidean distance. However, identifying optimal weights for similarity metric becomes a bottleneck because it involves performing a constrained exhaustive search. As a result, only a few predictors were selected and optimized in previous AnEn studies. A new machine learning similarity metric is proposed to improve the theoretical framework on how weather analogs are identified. First, a deep learning network is trained to generate latent features using all the temporal multivariate input predictors. Analogs are then selected in this latent space, rather than the original predictor space. The proposed method does not require prior predictor selection and an exhaustive search, thus presenting a significant computational benefit and scalability. It is tested for surface wind speed and solar irradiance forecasts in Pennsylvania from 2017 to 2019. Results show that the proposed method is capable of handling a large number of predictors, and it outperforms the original similarity metric in RMSE, bias, and CRPS. Since the data-driven transformation network is trained using the historical record, the proposed method has been found to be more flexible for searching through a longer record.
-
Working towards improved neuromyoelectric control of dexterous prosthetic hands, we explored how differences in training paradigms affect the subsequent online performance of two different motor-decode algorithms. Participants included two intact subjects and one participant who had undergone a recent transradial amputation after complex regional pain syndrome (CRPS) and multi-year disuse of the affected hand. During algorithm training sessions, participants actively mimicked hand movements appearing on a computer monitor. We varied both the duration of the hold-time (0.1 s or 5 s) at the end-point of each of six different digit and wrist movements, and the order in which the training movements were presented (random or sequential). We quantified the impact of these variations on two different motordecode algorithms, both having proportional, six-degree-offreedom (DOF) control: a modified Kalman filter (MKF) previously reported by this group, and a new approach - a convolutional neural network (CNN). Results showed that increasing the hold-time in the training set improved run-time performance. By contrast, presenting training movements in either random or sequential order had a variable and relatively modest effect on performance. The relative performance of the two decode algorithms varied according to the performance metric. This work represents the first-ever amputee use of a CNN for real-time, proportional six-DOF control of a prosthetic hand. Also novel was the testing of implanted high-channelcount devices for neuromyoelectric control shortly after amputation, following CRPS and long-term hand disuse. This work identifies key factors in the training of decode algorithms that improve their subsequent run-time performance.more » « less
-
Abstract We present a case study of solar flare forecasting by means of metadata feature time series, by treating it as a prominent class-imbalance and temporally coherent problem. Taking full advantage of pre-flare time series in solar active regions is made possible via the Space Weather Analytics for Solar Flares (SWAN-SF) benchmark data set, a partitioned collection of multivariate time series of active region properties comprising 4075 regions and spanning over 9 yr of the Solar Dynamics Observatory period of operations. We showcase the general concept of temporal coherence triggered by the demand of continuity in time series forecasting and show that lack of proper understanding of this effect may spuriously enhance models’ performance. We further address another well-known challenge in rare-event prediction, namely, the class-imbalance issue. The SWAN-SF is an appropriate data set for this, with a 60:1 imbalance ratio for GOES M- and X-class flares and an 800:1 imbalance ratio for X-class flares against flare-quiet instances. We revisit the main remedies for these challenges and present several experiments to illustrate the exact impact that each of these remedies may have on performance. Moreover, we acknowledge that some basic data manipulation tasks such as data normalization and cross validation may also impact the performance; we discuss these problems as well. In this framework we also review the primary advantages and disadvantages of using true skill statistic and Heidke skill score, two widely used performance verification metrics for the flare-forecasting task. In conclusion, we show and advocate for the benefits of time series versus point-in-time forecasting, provided that the above challenges are measurably and quantitatively addressed.more » « less
-
In this work, we tackle two widespread challenges in real applications for time series forecasting that have been largely understudied: distribution shifts and missing data. We propose SpectraNet, a novel multivariate time-series forecasting model that dynamically infers a latent space spectral decomposition to capture current temporal dynamics and correlations on the recent observed history. A Convolution Neural Network maps the learned representation by sequentially mixing its components and refining the output. Our proposed approach can simultaneously produce forecasts and interpolate past observations and can, therefore, greatly simplify production systems by unifying imputation and forecasting tasks into a single model. SpectraNet achieves SoTA performance simultaneously on both tasks on five benchmark datasets, compared to forecasting and imputation models, with up to 92% fewer parameters and comparable training times. On settings with up to 80% missing data, SpectraNet has average performance improvements of almost 50% over the second-best alternative.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1080/03050629.2022.2036987
