-
Representation Learning With Hidden Unit Clustering For Low Resource Speech Applications
Authors:
Varun Krishna,
Tarun Sai,
Sriram Ganapathy
Abstract:
The representation learning of speech, without textual resources, is an area of significant interest for many low resource speech applications. In this paper, we describe an approach to self-supervised representation learning from raw audio using a hidden unit clustering (HUC) framework. The input to the model consists of audio samples that are windowed and processed with 1-D convolutional layers.…
▽ More
The representation learning of speech, without textual resources, is an area of significant interest for many low resource speech applications. In this paper, we describe an approach to self-supervised representation learning from raw audio using a hidden unit clustering (HUC) framework. The input to the model consists of audio samples that are windowed and processed with 1-D convolutional layers. The learned "time-frequency" representations from the convolutional neural network (CNN) module are further processed with long short term memory (LSTM) layers which generate a contextual vector representation for every windowed segment. The HUC framework, allowing the categorization of the representations into a small number of phoneme-like units, is used to train the model for learning semantically rich speech representations. The targets consist of phoneme-like pseudo labels for each audio segment and these are generated with an iterative k-means algorithm. We explore techniques that improve the speaker invariance of the learned representations and illustrate the effectiveness of the proposed approach on two settings, i) completely unsupervised speech applications on the sub-tasks described as part of the ZeroSpeech 2021 challenge and ii) semi-supervised automatic speech recognition (ASR) applications on the TIMIT dataset and on the GramVaani challenge Hindi dataset. In these experiments, we achieve state-of-art results for various ZeroSpeech tasks. Further, on the ASR experiments, the HUC representations are shown to improve significantly over other established benchmarks based on Wav2vec, HuBERT and Best-RQ.
△ Less
Submitted 14 July, 2023;
originally announced July 2023.
-
Frequency and Amplitude Optimizations for Magnetic Particle Spectroscopy Applications
Authors:
Vinit Kumar Chugh,
Arturo di Girolamo,
Venkatramana D. Krishna,
Kai Wu,
Maxim C-J Cheeran,
Jian-** Wang
Abstract:
Nowadays, there is a growing interest in the field of magnetic particle spectroscopy (MPS)-based bioassays. MPS monitors the dynamic magnetic response of surface-functionalized magnetic nanoparticles (MNPs) upon excitation by an alternating magnetic field (AMF) to detect various target analytes. This technology has flourished in the past decade due to its low cost, low background magnetic noise in…
▽ More
Nowadays, there is a growing interest in the field of magnetic particle spectroscopy (MPS)-based bioassays. MPS monitors the dynamic magnetic response of surface-functionalized magnetic nanoparticles (MNPs) upon excitation by an alternating magnetic field (AMF) to detect various target analytes. This technology has flourished in the past decade due to its low cost, low background magnetic noise interference from biomatrix, and fast response time. A large number of MPS variants have been reported by different groups around the world, with applications ranging from disease diagnosis to foodborne pathogen detection, and virus detection. However, there is an urgent need for guidance on how to optimize the sensitivity of MPS detection by choosing different types of MNPs, AMF modalities, and MPS assay strategies (i.e., volume- and surface-based assays). In this work, we systematically study the effect of AMF frequencies and amplitudes on the responses of single- and multi-core MNPs under two extreme conditions, namely, the bound and unbound states. Our results show that some modalities such as dual-frequency MPS utilizing multicore MNPs are more suitable for surface-based bioassay applications, whereas, single-frequency MPS systems using single- or multi-core MNPs are better suited for volumetric bioassay applications. Furthermore, the bioassay sensitivities for these modalities can be further improved by careful selection of AMF frequencies and amplitudes.
△ Less
Submitted 10 November, 2022;
originally announced November 2022.
-
Magnetic Particle Spectroscopy (MPS) with One-stage Lock-in Implementation for Magnetic Bioassays with Improved Sensitivities
Authors:
Vinit Kumar Chugh,
Kai Wu,
Venkatramana D. Krishna,
Arturo di Girolamo,
Robert P. Bloom,
Yongqiang Andrew Wang,
Renata Saha,
Shuang Liang,
Maxim C-J Cheeran,
Jian-** Wang
Abstract:
In recent years, magnetic particle spectroscopy (MPS) has become a highly sensitive and versatile sensing technique for quantitative bioassays. It relies on the dynamic magnetic responses of magnetic nanoparticles (MNPs) for the detection of target analytes in liquid phase. There are many research studies reporting the application of MPS for detecting a variety of analytes including viruses, toxin…
▽ More
In recent years, magnetic particle spectroscopy (MPS) has become a highly sensitive and versatile sensing technique for quantitative bioassays. It relies on the dynamic magnetic responses of magnetic nanoparticles (MNPs) for the detection of target analytes in liquid phase. There are many research studies reporting the application of MPS for detecting a variety of analytes including viruses, toxins, and nucleic acids, etc. Herein, we report a modified version of MPS platform with the addition of a one-stage lock-in design to remove the feedthrough signals induced by external driving magnetic fields, thus capturing only MNP responses for improved system sensitivity. This one-stage lock-in MPS system is able to detect as low as 781 ng multi-core Nanomag50 iron oxide MNPs (micromod Partikeltechnologie GmbH) and 78 ng single-core SHB30 iron oxide MNPs (Ocean NanoTech). In addition, using a streptavidin-biotin binding system as a proof-of-concept, we show that these single-core SHB30 MNPs can be used for Brownian relaxation-based bioassays while the multi-core Nanomag50 cannot be used. The effects of MNP amount on the concentration dependent response profiles for detecting streptavidin was also investigated. Results show that by using lower concentration/amount of MNPs, concentration-response curves shift to lower concentration/amount of target analytes. This lower concentrationresponse indicates the possibility of improved bioassay sensitivities by using lower amounts of MNPs.
△ Less
Submitted 26 May, 2021;
originally announced May 2021.
-
GloFlow: Global Image Alignment for Creation of Whole Slide Images for Pathology from Video
Authors:
Viswesh Krishna,
Anirudh Joshi,
Philip L. Bulterys,
Eric Yang,
Andrew Y. Ng,
Pranav Rajpurkar
Abstract:
The application of deep learning to pathology assumes the existence of digital whole slide images of pathology slides. However, slide digitization is bottlenecked by the high cost of precise motor stages in slide scanners that are needed for position information used for slide stitching. We propose GloFlow, a two-stage method for creating a whole slide image using optical flow-based image registra…
▽ More
The application of deep learning to pathology assumes the existence of digital whole slide images of pathology slides. However, slide digitization is bottlenecked by the high cost of precise motor stages in slide scanners that are needed for position information used for slide stitching. We propose GloFlow, a two-stage method for creating a whole slide image using optical flow-based image registration with global alignment using a computationally tractable graph-pruning approach. In the first stage, we train an optical flow predictor to predict pairwise translations between successive video frames to approximate a stitch. In the second stage, this approximate stitch is used to create a neighborhood graph to produce a corrected stitch. On a simulated dataset of video scans of WSIs, we find that our method outperforms known approaches to slide-stitching, and stitches WSIs resembling those produced by slide scanners.
△ Less
Submitted 12 November, 2020; v1 submitted 28 October, 2020;
originally announced October 2020.
-
Data-Driven Evaluation of Building Demand Response Capacity
Authors:
Deokwoo Jung,
Varun Badrinath Krishna,
William Temple,
David K. Y. Yau
Abstract:
Before a building can participate in a demand response program, its facility managers must characterize the site's ability to reduce load. Today, this is often done through manual audit processes and prototypical control strategies. In this paper, we propose a new approach to estimate a building's demand response capacity using detailed data from various sensors installed in a building. We derive…
▽ More
Before a building can participate in a demand response program, its facility managers must characterize the site's ability to reduce load. Today, this is often done through manual audit processes and prototypical control strategies. In this paper, we propose a new approach to estimate a building's demand response capacity using detailed data from various sensors installed in a building. We derive a formula for a probabilistic measure that characterizes various tradeoffs between the available demand response capacity and the confidence level associated with that curtailment under the constraints of building occupant comfort level (or utility). Then, we develop a data-driven framework to associate observed or projected building energy consumption with a particular set of rules learned from a large sensor dataset. We apply this methodology using testbeds in two buildings in Singapore: a unique net-zero energy building and a modern commercial office building. Our experimental results identify key control parameters and provide insight into the available demand response strategies at each site.
△ Less
Submitted 9 February, 2016;
originally announced February 2016.
