A5076669165- Advanced Neural Network Applications
- Coronary Interventions and Diagnostics
- Cardiac Imaging and Diagnostics
- Lung Cancer Treatments and Mutations
- Acute Myocardial Infarction Research
- Lung Cancer Research Studies
- Peripheral Artery Disease Management
- Advanced Image and Video Retrieval Techniques
- Brain Tumor Detection and Classification
- CCD and CMOS Imaging Sensors
- Cardiac Valve Diseases and Treatments
- Neural Networks and Applications
- Image Enhancement Techniques
- Data Management and Algorithms
- Erythropoietin and Anemia Treatment
- Colorectal Cancer Treatments and Studies
- Speech Recognition and Synthesis
- Pancreatic and Hepatic Oncology Research
- Model Reduction and Neural Networks
- Context-Aware Activity Recognition Systems
- Lung Cancer Diagnosis and Treatment
- Web Data Mining and Analysis
- Advanced Database Systems and Queries
- Semantic Web and Ontologies
- Algorithms and Data Compression
National University of Singapore
2024
Advanced Micro Devices (Canada)
2023-2024
Xilinx (Ireland)
2021
Xilinx (United States)
2018
University of Kaiserslautern
2018
University of Campania "Luigi Vanvitelli"
2017
Politecnico di Milano
2016-2017
Ospedale Garibaldi
2016
University of Catania
2016
Azienda Ospedaliera San Giovanni Addolorata
2011-2014
Unresponsiveness to erythropoiesis-stimulating agents, occurring in 30% 50% of patients, is a major limitation the treatment chemotherapy-related anemia. We have prospectively evaluated whether intravenous iron can increase proportion patients with anemia who respond darbepoetin.Between December 2004 and February 2006, 149 lung, gynecologic, breast, colorectal cancers >or= 12 weeks planned chemotherapy were enrolled from 33 institutions. Patients required hemoglobin <or= 11 g/L no absolute...
The purpose of this single centre registry is to assess safety and feasibility the frequency domain optical coherence tomography (FD-OCT) system during coronary interventions.Ninety patients with unstable or stable artery disease were included in study. OCT imaging was performed a first group 40 (group 1), evaluate ambiguous/intermediate lesions (24 1 had also done post-PCI, for assessment stent deployment); second 50 2), address adequacy deployment. Therefore, 74 underwent FD-OCT after...
It is well known that many types of artificial neural networks, including recurrent can achieve a high classification accuracy even with low-precision weights and activations. The reduction in precision generally yields much more efficient hardware implementations regards to cost, memory requirements, energy, achievable throughput. In this paper, we present the first systematic exploration design space as function for Bidirectional Long Short-Term Memory (BiLSTM) network. Specifically,...
Background— Intravascular ultrasound (IVUS) studies have shown that a mechanism of plaque compression/embolization contributes toward the poststenting increase in lumen area. The aim this IVUS study was to compare mechanisms enlargement after coronary stenting 54 consecutive patients with unstable angina (UA) (group 1) and 56 stable 2) verify whether embolization plays major role former. Methods Results— Both groups underwent assessment (speed, 0.5 mm/sec) before intervention stent...
Schema mapping algorithms rely on value correspondences - i.e., among semantically related attributes to produce complex transformations data sources. These are either manually specified or suggested by separate modules called schema matchers. The quality of mappings produced a generation tool strongly depends the input correspondences. In this paper, we introduce Spicy system, novel approach problem verifying mappings. is based three-layer architecture, in which matching module used provide...
Glucagon-like peptide-1 (GLP-1) is a gut L-cell hormone that enhances glucose-stimulated insulin secretion. Several approaches prevent GLP-1 degradation or activate the receptor are being used to treat type 2 diabetes mellitus (T2DM) patients. In T2DM, secretion has been suggested be impaired, and this defect appears consequence rather than cause of impaired glucose homeostasis. However, although defective correlated with resistance, little known about direct effects chronic high...
Abstract The last decade brought significant advances in automatic speech recognition (ASR) thanks to the evolution of deep learning methods. ASR systems evolved from pipeline-based systems, that modeled hand-crafted features with probabilistic frameworks and generated phone posteriors, end-to-end (E2E) translate raw waveform directly into words using one neural network (DNN). transcription accuracy greatly increased, leading technology being integrated many commercial applications. However,...
Deep Neural Network (DNN) inference based on quantized narrow-precision integer data represents a promising research direction toward efficient deep learning computations edge and mobile devices. On one side, recent progress of Quantization-Aware Training (QAT) frameworks aimed at improving the accuracy extremely DNNs allows achieving results close to Floating-Point 32 (FP32), provides high flexibility concerning sizes selection. Unfortunately, current Central Processing Unit (CPU)...
Our research explores new forms of technology-enhanced interventions for children with Intellectual Disability (ID). The paper presents an innovative smart space called Magic K-Room, which has been designed in cooperation ID specialists and provides multisensory stimuli exploiting full-body interaction various kinds objects ambient features. Using brain signals acquired through a wearable EEG headset, the is responsive to particular child's levels relaxation attention, automatically adapting...
Efficient machine learning implementations optimized for inference in hardware have wide-ranging benefits, depending on the application, from lower latency to higher data throughput and reduced energy consumption. Two popular techniques reducing computation neural networks are pruning, removing insignificant synapses, quantization, precision of calculations. In this work, we explore interplay between pruning quantization during training ultra low applications targeting high physics use...
We present extensions to the Open Neural Network Exchange (ONNX) intermediate representation format represent arbitrary-precision quantized neural networks. first introduce support for low precision quantization in existing ONNX-based formats by leveraging integer clipping, resulting two new backward-compatible variants: operator with clipping and quantize-clip-dequantize (QCDQ) format. then a novel higher-level ONNX called (QONNX) that introduces three operators -- Quant, BipolarQuant,...
Deep neural networks (DNNs) are penetrating into a broad spectrum of applications and replacing manual algorithmic implementations, including the radio frequency communications domain with classical signal processing algorithms. However, high throughput (gigasamples per second) low latency requirements this application pose significant hurdle for adopting computationally demanding DNNs. In article, we explore highly specialized DNN inference accelerator approaches on field-programmable gate...
We introduce the Spicy system, a novel approach to problem of automatically selecting best mappings among two data sources. Known schema mapping algorithms rely on value correspondences -- i.e. semantically related attributes produce complex transformations brings together matching and generation tools further automate this process. A key observation, here, is that quality strongly influenced by input correspondences. To address problem, adopts three-layer architecture, in which module used...
The optimal management of unresectable locally advanced non-small-cell lung cancer in older patients has not been defined to date. present phase II study was planned evaluate the activity and safety platinum-based induction chemotherapy followed by concurrent chemoradiotherapy elderly with cancer. Patients received two cycles paclitaxel (175 mg/m2) carboplatin (area under curve: 5) day 1, every 3 weeks. Chemoradiotherapy (thoracic radiation therapy) initiated on 42 consisted 1.8 Gy daily,...
We introduce a quantization-aware training algorithm that guarantees avoiding numerical overflow when reducing the precision of accumulators during inference. leverage weight normalization as means constraining parameters using accumulator bit width bounds we derive. evaluate our across multiple quantized models train for different tasks, showing approach can reduce while maintaining model accuracy with respect to floating-point baseline. then show this reduction translates increased design...
We present accumulator-aware quantization (A2Q), a novel weight method designed to train quantized neural networks (QNNs) avoid overflow when using low-precision accumulators during inference. A2Q introduces unique formulation inspired by normalization that constrains the ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> -norm of model weights according accumulator bit width bounds we derive. Thus, in training QNNs for accumulation, also...