A5000408509- VLSI and Analog Circuit Testing
- Speech and Audio Processing
- Music and Audio Processing
- Advanced Data Processing Techniques
- Advanced Neural Network Applications
- Radiation Effects in Electronics
- Cryptography and Residue Arithmetic
- Neural Networks and Applications
- Cybersecurity and Information Systems
- Industrial Vision Systems and Defect Detection
- Low-power high-performance VLSI design
- Radiomics and Machine Learning in Medical Imaging
- Domain Adaptation and Few-Shot Learning
- VLSI and FPGA Design Techniques
- COVID-19 diagnosis using AI
- Speech Recognition and Synthesis
- Remote Sensing and LiDAR Applications
- Remote-Sensing Image Classification
- 3D Surveying and Cultural Heritage
- Advanced X-ray and CT Imaging
- Numerical Methods and Algorithms
- Advanced Computational Techniques in Science and Engineering
- 3D IC and TSV technologies
- Chaos-based Image/Signal Encryption
- Integrated Circuits and Semiconductor Failure Analysis
Institute for Design Problems in Microelectronics
2014-2024
Tencent (China)
2024
Mitsui (Japan)
2024
Korea University
2023-2024
Centro Universitário de João Pessoa
2023-2024
Central Conservatory of Music
2023-2024
Université de Montréal
2022
Russian Academy of Sciences
2022
National Institutes of Health
2022
Aarhus University
2022
This paper summarizes the music demixing (MDX) track of Sound Demixing Challenge (SDX'23).We provide a summary challenge setup and introduce task robust source separation (MSS), i.e., training MSS models in presence errors data.We propose formalization that can occur design dataset for systems two new datasets simulate such errors: SDXDB23_LabelNoise SDXDB23_Bleeding 1 .We describe methods achieved highest scores competition.Moreover, we present direct comparison with previous edition (the...
This paper summarizes the cinematic demixing (CDX) track of Sound Demixing Challenge 2023 (SDX'23). We provide a comprehensive summary challenge setup, detailing structure competition and datasets used. Especially, we detail CDXDB23, new hidden dataset constructed from real movies that was used to rank submissions. The also offers insights into most successful approaches employed by participants. Compared cocktail-fork baseline, best-performing system trained exclusively on simulated Divide...
Automatic classification of sound commands is becoming increasingly important, especially for embedded and mobile devices. Many these devices contain both microphones cameras. The manufacturers that develop produce them would like to use the same methodology image tasks. It's possible achieve by representing as images, then convolutional neural networks when classifying images well sounds. In this research, we tried several approaches problem applied in TensorFlow Speech Recognition...
Abstract While spatial proteomics by fluorescence imaging has quickly become an essential discovery tool for researchers, fast and scalable methods to classify embed single-cell protein distributions in such images are lacking. Here, we present the design analysis of results from competition Human Protein Atlas – Single-Cell Classification hosted on Kaggle platform. This represents a crowd-sourced develop machine learning models trained limited annotations label patterns fluorescent images....
In this paper, we present PAGR (Python Alpha Global Routing) – a solution to the global routing problem in physical synthesis based on data from ISPD 2024 contest. Our constructs weighted graph and builds Steiner tree. To accelerate tree search, propose technique for size minimization by reducing input 3D matrix. This method slightly decreases result quality but finds solutions 2–10 times faster. We also detail our methods parallel calculations computing edge weights. Experimental results...
This article describes a brief history of the development CAD system for integrated circuit design in Russia. The background beginning import-independent Russian digital "Obsidian" is presented. What tasks were set developers and which results have been achieved up to present. large tests obtained by compare this with commercial tools are Further directions circuits determined separate stages route.
Modern computational tasks are often required to not only guarantee predefined accuracy, but get the result fast. Optimizing calculations using floating point numbers, as opposed integers, is a non-trivial task. For this reason, there need explore new ways improve such operations. This paper presents analysis and comparison of various formats - float, posit bfloat. One promising areas in which problem values can be considered most acute neural networks. That why we pay special attention...
Modern mobile neural networks with a reduced number of weights and parameters do good job image classification tasks, but even they may be too complex to implemented in an FPGA for video processing tasks. The article proposes network architecture the practical task recognizing images from camera, which has several advantages terms speed. This is achieved by reducing weights, moving floating-point fixed-point arithmetic, due hardware-level optimizations associated storing blocks, shift...
Over a finite field transformation similar to discrete Fourier transform can be defined that efficiently implemented using fast algorithms. One of the main applications such is calculation convolutions long sequences integers by means integer arithmetic. In this paper method implementation two vectors convolution modular arithmetic with Proth-type modulo considered. A device performing cyclic was created. comparison binary analogues carried out.
The last decades have witnessed rapid IoT technologies development, which provided ubiquitous human-computer interactions. Building intelligent systems of various types, among emotion recognition systems, is important challenge nowadays. Especially pressing problem to build a real-time portable system can be embedded in low performance hardware. We propose high accuracy system, deployed on single board Raspberry Pi computer perform 4 facial expressions: neutral, angry, surprised, and happy....
Medical research has made tremendous progress in detecting various pathologies the human body. There is still problem of speed process, and lack a sufficient number trained professionals this field. Detection prostate cancer, particular, without surgery very labor- intensive process. A neural network-based machine learning algorithm been proposed to solve problem, making it possible see suspected areas lesions organ. In study, comprehensive analysis TRUS image processing approaches was...
Music demixing is the task of separating different tracks from given single audio signal into components, such as drums, bass, and vocals rest accompaniment. Separation sources useful for a range areas, including entertainment hearing aids. In this paper, we introduce two new benchmarks sound source separation tasks compare popular models demixing, well their ensembles, on these benchmarks. For models' assessments, provide leaderboard at https://mvsep.com/quality_checker/, giving comparison...
This article provides an overview of the current state research in field reliability combinational circuits. The developed methods for circuits generation and their subsequent implementation form software are described. implemented algorithms were used to generate structured Verilog format; generated converted using Nangate open standard cell library. For schemes, parameters calculated stored a conversion into dataset csv format. Using software, was generated; it can be conduct on predicting...
Timing margining is a key component of timing sign-off. Insufficient margin can lead to silicon failure and excessive pessimistic will entail unnecessary design optimization effort. intended cover the uncertainty in clock arrival times skews arising from within-die process variations. In highly scaled technologies, increased variations tend enforce an overestimation margins making it difficult for designs achieve target performance. this paper, we present more efficient methodology account...
Approximate synthesis is a modern trend in the field of logic synthesis, which makes it possible to obtain much more compact, high-speed and reliable solutions due weakening requirements for accuracy implemented functions. For number applications, small distortions at outputs can be than acceptable, improvement characteristics powerful argument favor this method. We propose new approach approximate combinational logic, built upon solving regression problem with use iterative methods based on...
In this paper convolutional neural networks for video stream processing using hardware with limited computing resource are studied.To solve problems of performance, we propose to move from software implementation, and also fixed-point floating point calculations.A set methods network design was proposed aimed at maximal performance accuracy in case fixed calculations.As the example, describe solving a specific practical task.We show how existing datasets can be adapted another task different...