RESUMEN
The record-breaking performance of deep neural networks (DNNs) comes with heavy parameter budgets, which leads to external dynamic random access memory (DRAM) for storage. The prohibitive energy of DRAM accesses makes it nontrivial for DNN deployment on resource-constrained devices, calling for minimizing the movements of weights and data in order to improve the energy efficiency. Driven by this critical bottleneck, we present SmartDeal, a hardware-friendly algorithm framework to trade higher-cost memory storage/access for lower-cost computation, in order to aggressively boost the storage and energy efficiency, for both DNN inference and training. The core technique of SmartDeal is a novel DNN weight matrix decomposition framework with respective structural constraints on each matrix factor, carefully crafted to unleash the hardware-aware efficiency potential. Specifically, we decompose each weight tensor as the product of a small basis matrix and a large structurally sparse coefficient matrix whose nonzero elements are readily quantized to the power-of-2. The resulting sparse and readily quantized DNNs enjoy greatly reduced energy consumption in data movement as well as weight storage, while incurring minimal overhead to recover the original weights thanks to the required sparse bit-operations and cost-favorable computations. Beyond inference, we take another leap to embrace energy-efficient training, by introducing several customized techniques to address the unique roadblocks arising in training while preserving the SmartDeal structures. We also design a dedicated hardware accelerator to fully utilize the new weight structure to improve the real energy efficiency and latency performance. We conduct experiments on both vision and language tasks, with nine models, four datasets, and three settings (inference-only, adaptation, and fine-tuning). Our extensive results show that 1) being applied to inference, SmartDeal achieves up to 2.44× improvement in energy efficiency as evaluated using real hardware implementations and 2) being applied to training, SmartDeal can lead to 10.56× and 4.48× reduction in the storage and the training energy cost, respectively, with usually negligible accuracy loss, compared to state-of-the-art training baselines. Our source codes are available at: https://github.com/VITA-Group/SmartDeal.
RESUMEN
There exists a gap in terms of the signals provided by pacemakers (i.e., intracardiac electrogram (EGM)) and the signals doctors use (i.e., 12-lead electrocardiogram (ECG)) to diagnose abnormal rhythms. Therefore, the former, even if remotely transmitted, are not sufficient for doctors to provide a precise diagnosis, let alone make a timely intervention. To close this gap and make a heuristic step towards real-time critical intervention in instant response to irregular and infrequent ventricular rhythms, we propose a new framework dubbed RT-RCG to automatically search for (1) efficient Deep Neural Network (DNN) structures and then (2) corresponding accelerators, to enable Real-Time and high-quality Reconstruction of ECG signals from EGM signals. Specifically, RT-RCG proposes a new DNN search space tailored for ECG reconstruction from EGM signals, and incorporates a differentiable acceleration search (DAS) engine to efficiently navigate over the large and discrete accelerator design space to generate optimized accelerators. Extensive experiments and ablation studies under various settings consistently validate the effectiveness of our RT-RCG. To the best of our knowledge, RT-RCG is the first to leverage neural architecture search (NAS) to simultaneously tackle both reconstruction efficacy and efficiency.
RESUMEN
PURPOSE: The aim of this study was to develop a pretherapy PET/CT-based prediction model for treatment response to ibrutinib in lymphoma patients. PATIENTS AND METHODS: One hundred sixty-nine lymphoma patients with 2441 lesions were studied retrospectively. All eligible lymphomas on pretherapy 18F-FDG PET images were contoured and segmented for radiomic analysis. Lesion- and patient-based responsiveness to ibrutinib was determined retrospectively using the Lugano classification. PET radiomic features were extracted. A radiomic model was built to predict ibrutinib response. The prognostic significance of the radiomic model was evaluated independently in a test cohort and compared with conventional PET metrics: SUVmax, metabolic tumor volume, and total lesion glycolysis. RESULTS: The radiomic model had an area under the receiver operating characteristic curve (ROC AUC) of 0.860 (sensitivity, 92.9%, specificity, 81.4%; P < 0.001) for predicting response to ibrutinib, outperforming the SUVmax (ROC AUC, 0.519; P = 0.823), metabolic tumor volume (ROC AUC, 0.579; P = 0.412), total lesion glycolysis (ROC AUC, 0.576; P = 0.199), and a composite model built using all 3 (ROC AUC, 0.562; P = 0.046). The radiomic model increased the probability of accurately predicting ibrutinib-responsive lesions from 84.8% (pretest) to 96.5% (posttest). At the patient level, the model's performance (ROC AUC = 0.811; P = 0.007) was superior to that of conventional PET metrics. Furthermore, the radiomic model showed robustness when validated in treatment subgroups: first (ROC AUC, 0.916; P < 0.001) versus second or greater (ROC AUC, 0.842; P < 0.001) line of defense and single treatment (ROC AUC, 0.931; P < 0.001) versus multiple treatments (ROC AUC, 0.824; P < 0.001). CONCLUSIONS: We developed and validated a pretherapy PET-based radiomic model to predict response to treatment with ibrutinib in a diverse cohort of lymphoma patients.
