This paper presents a novel zero-shot learning approach towards personalized speech enhancement through the use of a sparsely active ensemble model. Optimizing speech denoising systems towards a particular test-time speaker can improve performance and reduce run-time complexity. However, test-time model adaptation may be challenging if collecting data from the test-time speaker is not possible. To this end, we propose using an ensemble model wherein each specialist module denoises noisy utterances from a distinct partition of training set speakers. The gating module inexpensively estimates test-time speaker characteristics in the form of an embedding vector and selects the most appropriate specialist module for denoising the test signal. Grouping the training set speakers into non-overlapping semantically similar groups is non-trivial and ill-defined. To do this, we first train a Siamese network using noisy speech pairs to maximize or minimize the similarity of its output vectors depending on whether the utterances derive from the same speaker or not. Next, we perform k-means clustering on the latent space formed by the averaged embedding vectors per training set speaker. In this way, we designate speaker groups and train specialist modules optimized around partitions of the complete training set. Our experiments show that ensemble models made up of low-capacity specialists can outperform high-capacity generalist models with greater efficiency and improved adaptation towards unseen test-time speakers.
more »
« less
Diverse Knowledge Distillation (DKD): A Solution for Improving The Robustness of Ensemble Models Against Adversarial Attacks
This paper proposes an ensemble learning model that is resistant to adversarial attacks. To build resilience, we introduced a training process where each member learns a radically distinct latent space. Member models are added one at a time to the ensemble. Simultaneously, the loss function is regulated by a reverse knowledge distillation, forcing the new member to learn different features and map to a latent space safely distanced from those of existing members. We assessed the security and performance of the proposed solution on image classification tasks using CIFAR10 and MNIST datasets and showed security and performance improvement compared to the state of the art defense methods.
more »
« less
- NSF-PAR ID:
- 10298697
- Date Published:
- Journal Name:
- 22nd International Symposium on Quality Electronic Design (ISQED)
- Page Range / eLocation ID:
- 319 to 324
- 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.
-
more » « less
Abstract Stochastic model error schemes, such as the stochastic perturbed parameterization tendencies (SPPT) and independent SPPT (iSPPT) schemes, have become an increasingly accepted method to represent model error associated with uncertain subgrid-scale processes in ensemble prediction systems (EPSs). While much of the current literature focuses on the effects of these schemes on forecast skill, this research examines the physical processes by which iSPPT perturbations to the microphysics parameterization scheme yield variability in ensemble rainfall forecasts. Members of three 120-member Weather Research and Forecasting (WRF) Model ensemble case studies, including two distinct heavy rain events over Taiwan and one over the northeastern United States, are ranked according to an area-averaged accumulated rainfall metric in order to highlight differences between high- and low-precipitation forecasts. In each case, high-precipitation members are characterized by a damping of the microphysics water vapor and temperature tendencies over the region of heaviest rainfall, while the opposite is true for low-precipitation members. Physically, the perturbations to microphysics tendencies have the greatest impact at the cloud level and act to modify precipitation efficiency. To this end, the damping of tendencies in high-precipitation forecasts suppresses both the loss of water vapor due to condensation and the corresponding latent heat release, leading to grid-scale supersaturation. Conversely, amplified tendencies in low-precipitation forecasts yield both drying and increased positive buoyancy within clouds.
-
more » « less
Abstract Arctic Ocean warming and sea ice loss are closely linked to increased ocean heat transport (OHT) into the Arctic and changes in surface heat fluxes. To quantitatively assess their respective roles, we use the 100-member Community Earth System Model, version 2 (CESM2), Large Ensemble over the 1920–2100 period. We first examine the Arctic Ocean warming in a heat budget framework by calculating the contributions from heat exchanges with atmosphere and sea ice and OHT across the Arctic Ocean gateways. Then we quantify how much anomalous heat from the ocean directly translates to sea ice loss and how much is lost to the atmosphere. We find that Arctic Ocean warming is driven primarily by increased OHT through the Barents Sea Opening, with additional contributions from the Fram Strait and Bering Strait OHTs. These OHT changes are driven mainly by warmer inflowing water rather than changes in volume transports across the gateways. The Arctic Ocean warming driven by OHT is partially damped by increased heat loss through the sea surface. Although absorbed shortwave radiation increases due to reduced surface albedo, this increase is compensated by increasing upwelling longwave radiation and latent heat loss. We also explicitly calculate the contributions of ocean–ice and atmosphere–ice heat fluxes to sea ice heat budget changes. Throughout the entire twentieth century as well as the early twenty-first century, the atmosphere is the main contributor to ice heat gain in summer, though the ocean’s role is not negligible. Over time, the ocean progressively becomes the main heat source for the ice as the ocean warms.
Significance Statement Arctic Ocean warming and sea ice loss are closely linked to increased ocean heat transport (OHT) into the Arctic and changes in surface heat fluxes. Here we use 100 simulations from the same climate model to analyze future warming and sea ice loss. We find that Arctic Ocean warming is primarily driven by increased OHT through the Barents Sea Opening, though the Fram and Bering Straits are also important. This increased OHT is primarily due to warmer inflowing water rather than changing ocean currents. This ocean heat gain is partially compensated by heat loss through the sea surface. During the twentieth century and early twenty-first century, sea ice loss is mainly linked to heat transferred from the atmosphere; however, over time, the ocean progressively becomes the most important contributor.
-
more » « less
Abstract To assess the effect of uncertainties in solar wind driving on the predictions from the operational configuration of the Space Weather Modeling Framework, we have developed a nonparametric method for generating multiple possible realizations of the solar wind just upstream of the bow shock, based on observations near the first Lagrangian point. We have applied this method to the solar wind inputs at the upstream boundary of Space Weather Modeling Framework and have simulated the geomagnetic storm of 5 April 2010. We ran a 40‐member ensemble for this event and have used this ensemble to quantify the uncertainty in the predicted Sym‐H index and ground magnetic disturbances due to the uncertainty in the upstream boundary conditions. Both the ensemble mean and the unperturbed simulation tend to underpredict the magnitude of Sym‐H in the quiet interval before the storm and overpredict in the storm itself, consistent with previous work. The ensemble mean is a more accurate predictor of Sym‐H, improving the mean absolute error by nearly 2 nT for this interval and displaying a smaller bias. We also examine the uncertainty in predicted maxima in ground magnetic disturbances. The confidence intervals are typically narrow during periods where the predicted d
B H /dt is low. The confidence intervals are often much wider where the median prediction is for enhanced dB H /dt . The ensemble also allows us to identify intervals of activity that cannot be explained by uncertainty in the solar wind driver, driving further model improvements. This work demonstrates the feasibility and importance of ensemble modeling for space weather applications.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ISQED51717.2021.9424353
