Showing papers by "Indian Institute of Technology Bombay published in 2020"

GW190425: Observation of a Compact Binary Coalescence with Total Mass ∼ 3.4 M O

Abstract: On 2019 April 25, the LIGO Livingston detector observed a compact binary coalescence with signal-to-noise ratio 12.9. The Virgo detector was also taking data that did not contribute to detection due to a low signal-to-noise ratio, but were used for subsequent parameter estimation. The 90% credible intervals for the component masses range from to if we restrict the dimensionless component spin magnitudes to be smaller than 0.05). These mass parameters are consistent with the individual binary components being neutron stars. However, both the source-frame chirp mass and the total mass of this system are significantly larger than those of any other known binary neutron star (BNS) system. The possibility that one or both binary components of the system are black holes cannot be ruled out from gravitational-wave data. We discuss possible origins of the system based on its inconsistency with the known Galactic BNS population. Under the assumption that the signal was produced by a BNS coalescence, the local rate of neutron star mergers is updated to 250-2810.

GW190814: Gravitational Waves from the Coalescence of a 23 Solar Mass Black Hole with a 2.6 Solar Mass Compact Object

Abstract: We report the observation of a compact binary coalescence involving a 222–243 M ⊙ black hole and a compact object with a mass of 250–267 M ⊙ (all measurements quoted at the 90% credible level) The gravitational-wave signal, GW190814, was observed during LIGO's and Virgo's third observing run on 2019 August 14 at 21:10:39 UTC and has a signal-to-noise ratio of 25 in the three-detector network The source was localized to 185 deg2 at a distance of ${241}_{-45}^{+41}$ Mpc; no electromagnetic counterpart has been confirmed to date The source has the most unequal mass ratio yet measured with gravitational waves, ${0112}_{-0009}^{+0008}$, and its secondary component is either the lightest black hole or the heaviest neutron star ever discovered in a double compact-object system The dimensionless spin of the primary black hole is tightly constrained to ≤007 Tests of general relativity reveal no measurable deviations from the theory, and its prediction of higher-multipole emission is confirmed at high confidence We estimate a merger rate density of 1–23 Gpc−3 yr−1 for the new class of binary coalescence sources that GW190814 represents Astrophysical models predict that binaries with mass ratios similar to this event can form through several channels, but are unlikely to have formed in globular clusters However, the combination of mass ratio, component masses, and the inferred merger rate for this event challenges all current models of the formation and mass distribution of compact-object binaries

GW190521: A Binary Black Hole Merger with a Total Mass of 150 M

Abstract: On May 21, 2019 at 03:02:29 UTC Advanced LIGO and Advanced Virgo observed a short duration gravitational-wave signal, GW190521, with a three-detector network signal-to-noise ratio of 14.7, and an estimated false-alarm rate of 1 in 4900 yr using a search sensitive to generic transients. If GW190521 is from a quasicircular binary inspiral, then the detected signal is consistent with the merger of two black holes with masses of 85_{-14}^{+21} M_{⊙} and 66_{-18}^{+17} M_{⊙} (90% credible intervals). We infer that the primary black hole mass lies within the gap produced by (pulsational) pair-instability supernova processes, with only a 0.32% probability of being below 65 M_{⊙}. We calculate the mass of the remnant to be 142_{-16}^{+28} M_{⊙}, which can be considered an intermediate mass black hole (IMBH). The luminosity distance of the source is 5.3_{-2.6}^{+2.4} Gpc, corresponding to a redshift of 0.82_{-0.34}^{+0.28}. The inferred rate of mergers similar to GW190521 is 0.13_{-0.11}^{+0.30} Gpc^{-3} yr^{-1}.

GW190412: Observation of a binary-black-hole coalescence with asymmetric masses

Abstract: We report the observation of gravitational waves from a binary-black-hole coalescence during the first two weeks of LIGO’s and Virgo’s third observing run. The signal was recorded on April 12, 2019 at 05∶30∶44 UTC with a network signal-to-noise ratio of 19. The binary is different from observations during the first two observing runs most notably due to its asymmetric masses: a ∼30 M⊙ black hole merged with a ∼8 M⊙ black hole companion. The more massive black hole rotated with a dimensionless spin magnitude between 0.22 and 0.60 (90% probability). Asymmetric systems are predicted to emit gravitational waves with stronger contributions from higher multipoles, and indeed we find strong evidence for gravitational radiation beyond the leading quadrupolar order in the observed signal. A suite of tests performed on GW190412 indicates consistency with Einstein’s general theory of relativity. While the mass ratio of this system differs from all previous detections, we show that it is consistent with the population model of stellar binary black holes inferred from the first two observing runs.

Blockchain technology in supply chain operations: Applications, challenges and research opportunities

Abstract: Blockchain is a technology with unique combination of features such as decentralized structure, distributed notes and storage mechanism, consensus algorithm, smart contracting, and asymmetric encryption to ensure network security, transparency and visibility. Blockchain has immense potential to transform supply chain (SC) functions, from SC provenance, business process reengineering to security enhancement. More and more studies exploring the use of blockchain in SCs have appeared in recent years. In this paper, we consider a total of 178 articles and examine all the relevant research done in the field associated with the use of blockchain integration in SC operations. We highlight the corresponding opportunities, possible societal impacts, current state-of-the-art technologies along with major trends and challenges. We examine several industrial sectors such as shipping, manufacturing, automotive, aviation, finance, technology, energy, healthcare, agriculture and food, e-commerce, and education among others that can be successfully revamped with blockchain based technologies through enhanced visibility and business process management. A future research agenda is established which lays the solid foundation for further studies on this important emerging research area.

Properties and Astrophysical Implications of the 150 M Binary Black Hole Merger GW190521

Abstract: The gravitational-wave signal GW190521 is consistent with a binary black hole (BBH) merger source at redshift 0.8 with unusually high component masses, 85-14+21 M o˙ and 66-18+17 M o˙, compared to previously reported events, and shows mild evidence for spin-induced orbital precession. The primary falls in the mass gap predicted by (pulsational) pair-instability supernova theory, in the approximate range 65-120 M o˙. The probability that at least one of the black holes in GW190521 is in that range is 99.0%. The final mass of the merger (142-16+28 M o˙) classifies it as an intermediate-mass black hole. Under the assumption of a quasi-circular BBH coalescence, we detail the physical properties of GW190521's source binary and its post-merger remnant, including component masses and spin vectors. Three different waveform models, as well as direct comparison to numerical solutions of general relativity, yield consistent estimates of these properties. Tests of strong-field general relativity targeting the merger-ringdown stages of the coalescence indicate consistency of the observed signal with theoretical predictions. We estimate the merger rate of similar systems to be 0.13-0.11+0.30 Gpc-3 yr-1. We discuss the astrophysical implications of GW190521 for stellar collapse and for the possible formation of black holes in the pair-instability mass gap through various channels: via (multiple) stellar coalescences, or via hierarchical mergers of lower-mass black holes in star clusters or in active galactic nuclei. We find it to be unlikely that GW190521 is a strongly lensed signal of a lower-mass black hole binary merger. We also discuss more exotic possible sources for GW190521, including a highly eccentric black hole binary, or a primordial black hole binary.

α-Synuclein aggregation nucleates through liquid-liquid phase separation.

Abstract: α-Synuclein (α-Syn) aggregation and amyloid formation is directly linked with Parkinson's disease pathogenesis. However, the early events involved in this process remain unclear. Here, using the in vitro reconstitution and cellular model, we show that liquid-liquid phase separation of α-Syn precedes its aggregation. In particular, in vitro generated α-Syn liquid-like droplets eventually undergo a liquid-to-solid transition and form an amyloid hydrogel that contains oligomers and fibrillar species. Factors known to aggravate α-Syn aggregation, such as low pH, phosphomimetic substitution and familial Parkinson's disease mutations, also promote α-Syn liquid-liquid phase separation and its subsequent maturation. We further demonstrate α-Syn liquid-droplet formation in cells. These cellular α-Syn droplets eventually transform into perinuclear aggresomes, the process regulated by microtubules. This work provides detailed insights into the phase-separation behaviour of natively unstructured α-Syn and its conversion to a disease-associated aggregated state, which is highly relevant in Parkinson's disease pathogenesis.

A Multi-Organ Nucleus Segmentation Challenge

Abstract: Generalized nucleus segmentation techniques can contribute greatly to reducing the time to develop and validate visual biomarkers for new digital pathology datasets. We summarize the results of MoNuSeg 2018 Challenge whose objective was to develop generalizable nuclei segmentation techniques in digital pathology. The challenge was an official satellite event of the MICCAI 2018 conference in which 32 teams with more than 80 participants from geographically diverse institutes participated. Contestants were given a training set with 30 images from seven organs with annotations of 21,623 individual nuclei. A test dataset with 14 images taken from seven organs, including two organs that did not appear in the training set was released without annotations. Entries were evaluated based on average aggregated Jaccard index (AJI) on the test set to prioritize accurate instance segmentation as opposed to mere semantic segmentation. More than half the teams that completed the challenge outperformed a previous baseline. Among the trends observed that contributed to increased accuracy were the use of color normalization as well as heavy data augmentation. Additionally, fully convolutional networks inspired by variants of U-Net, FCN, and Mask-RCNN were popularly used, typically based on ResNet or VGG base architectures. Watershed segmentation on predicted semantic segmentation maps was a popular post-processing strategy. Several of the top techniques compared favorably to an individual human annotator and can be used with confidence for nuclear morphometrics.

Bayesian inference for compact binary coalescences with bilby: validation and application to the first LIGO–Virgo gravitational-wave transient catalogue

Abstract: Gravitational waves provide a unique tool for observational astronomy. While the first LIGO–Virgo catalogue of gravitational-wave transients (GWTC-1) contains 11 signals from black hole and neutron star binaries, the number of observations is increasing rapidly as detector sensitivity improves. To extract information from the observed signals, it is imperative to have fast, flexible, and scalable inference techniques. In a previous paper, we introduced bilby: a modular and user-friendly Bayesian inference library adapted to address the needs of gravitational-wave inference. In this work, we demonstrate that bilby produces reliable results for simulated gravitational-wave signals from compact binary mergers, and verify that it accurately reproduces results reported for the 11 GWTC-1 signals. Additionally, we provide configuration and output files for all analyses to allow for easy reproduction, modification, and future use. This work establishes that bilby is primed and ready to analyse the rapidly growing population of compact binary coalescence gravitational-wave signals.

A global anthropogenic emission inventory of atmospheric pollutants from sector- and fuel-specific sources (1970-2017): an application of the Community Emissions Data System (CEDS)

Abstract: . Global anthropogenic emission inventories remain vital for understanding the fate and transport of atmospheric pollution, as well as the resulting impacts on the environment, human health, and society. Rapid changes in today’s society require that these inventories provide contemporary estimates of multiple atmospheric pollutants with both source sector and fuel-type information to understand and effectively mitigate future impacts. To fill this need, we have updated the open-source Community Emissions Data System (CEDS) (Hoesly et al., 2019) to develop a new global emission inventory, CEDSGBD-MAPS. This inventory includes emissions of seven key atmospheric pollutants (NOx, CO, SO2, NH3, NMVOCs, BC, OC) over the time period from 1970–2017 and reports annual country-total emissions as a function of 11 anthropogenic sectors (agriculture, energy generation, industrial processes, transportation (on-road and non-road), residential, commercial, and other sectors (RCO), waste, solvent use, and international-shipping) and four fuel categories (total coal, solid biofuel, and the sum of liquid fuels and natural gas combustion, plus remaining process-level emissions). The CEDSGBD-MAPS inventory additionally includes global gridded (0.5°×0.5°) emission fluxes with monthly time resolution for each compound, sector, and fuel-type to facilitate their use in earth system models. CEDSGBD-MAPS utilizes updated activity data, updates to the core CEDS default calibration procedure, and modifications to the final procedures for emissions gridding and aggregation to retain sector and fuel-specific information. Relative to the previous CEDS data released for CMIP6 (Hoesly et al., 2018), these updates extend the emission estimates from 2014 to 2017 and improve the overall agreement between CEDS and two widely used global bottom-up emission inventories. The CEDSGBD-MAPS inventory provides the most contemporary global emission estimates to-date for these key atmospheric pollutants and is the first to provide global estimates for these species as a function of multiple fuel-types across multiple source sectors. Dominant sources of global NOx and SO2 emissions in 2017 include the combustion of oil, gas, and coal in the energy and industry sectors, as well as on-road transportation and international shipping for NOx. Dominant sources of global CO emissions in 2017 include on-road transportation and residential biofuel combustion. Dominant global sources of carbonaceous aerosol in 2017 include residential biofuel combustion, on-road transportation (BC only), as well as emissions from waste. Global emissions of NOx, SO2, CO, BC, and OC all peak in 2012 or earlier, with more recent emission reductions driven by large changes in emissions from China, North America, and Europe. In contrast, global emissions of NH3 and NMVOCs continuously increase between 1970 and 2017, with agriculture serving as a major source of global NH3 emissions and solvent use, energy, residential, and the on-road transport sectors as major sources of global NMVOCs. Due to similar development methods and underlying datasets, the CEDSGBD-MAPS emissions are expected to have consistent sources of uncertainty as other bottom-up inventories, including uncertainties in the underlying activity data and sector- and region-specific emission factors. The CEDSGBD-MAPS source code is publicly available online through GitHub: https://github.com/emcduffie/CEDS/tree/CEDS_GBD-MAPS. The CEDSGBD-MAPS emission inventory dataset (both annual country-total and global gridded files) is publicly available and registered under: https://doi.org/10.5281/zenodo.3754964 (McDuffie et al., 2020).

A guide to LIGO–Virgo detector noise and extraction of transient gravitational-wave signals

Abstract: The LIGO Scientific Collaboration and the Virgo Collaboration have cataloged eleven confidently detected gravitational-wave events during the first two observing runs of the advanced detector era. All eleven events were consistent with being from well-modeled mergers between compact stellar-mass objects: black holes or neutron stars. The data around the time of each of these events have been made publicly available through the gravitational-wave open science center. The entirety of the gravitational-wave strain data from the first and second observing runs have also now been made publicly available. There is considerable interest among the broad scientific community in understanding the data and methods used in the analyses. In this paper, we provide an overview of the detector noise properties and the data analysis techniques used to detect gravitational-wave signals and infer the source properties. We describe some of the checks that are performed to validate the analyses and results from the observations of gravitational-wave events. We also address concerns that have been raised about various properties of LIGO–Virgo detector noise and the correctness of our analyses as applied to the resulting data.

Hybrid ozonation process for industrial wastewater treatment: Principles and applications: A review

Abstract: Ozone is a strong oxidant and have been effectively used for the degradation and mineralization of organic pollutants. However, the increase in the toxicity and disposal of the recalcitrant organics standalone ozonation process is not effective and sustainable solution for the treatment of industrial wastewater containing recalcitrant. It is therefore necessary to provide a summary of success of hybrid ozonation process for industrial wastewater treatment along with the reaction mechanism for enhancing the molecular ozone reactivity. The paper presents a detailed review of hybrid ozonation process as a combination of two different techniques to enhance the hydroxyl radical formation thereby increasing the process efficiency. An extensive review on the mechanism and application of these hybrid ozonation processes for degradation, mineralization, detoxification of different organic pollutants present in the industrial wastewater is reported.

Likelihood of survival of coronavirus in a respiratory droplet deposited on a solid surface

Abstract: We predict and analyze the drying time of respiratory droplets from a COVID-19 infected subject, which is a crucial time to infect another subject. Drying of the droplet is predicted by using a diffusion-limited evaporation model for a sessile droplet placed on a partially wetted surface with a pinned contact line. The variation in droplet volume, contact angle, ambient temperature, and humidity are considered. We analyze the chances of the survival of the virus present in the droplet based on the lifetime of the droplets under several conditions and find that the chances of the survival of the virus are strongly affected by each of these parameters. The magnitude of shear stress inside the droplet computed using the model is not large enough to obliterate the virus. We also explore the relationship between the drying time of a droplet and the growth rate of the spread of COVID-19 in five different cities and find that they are weakly correlated.

A comprehensive review on recent preparation techniques of liposomes.

Abstract: Liposomes (or lipid vesicles) are a versatile platform as carriers for the delivery of the drugs and other macromolecules into human and animal bodies. Though the method of using liposomes has been known since 1960s, and major developments and commercialization of liposomal formulations took place in the late nineties (or early part of this century), newer methods of liposome synthesis and drug encapsulation continue to be an active area of research. With the developments in related fields, such as electrohydrodynamics and microfluidics, and a growing understanding of the mechanisms of lipid assembly from colloidal and intermolecular forces, liposome preparation techniques have been enriched and more predictable than before. This has led to better methods that can also scale at an industrial production level. In this review, we present several novel methods that were introduced over the last decade and compare their advantages over conventional methods. Researchers beginning to explore liposomal formulations will find this resource useful to give an overall direction for an appropriate choice of method. Where possible, we have also provided the known mechanisms behind the preparation methods.

A high-stringency blueprint of the human proteome

Abstract: The Human Proteome Organization (HUPO) launched the Human Proteome Project (HPP) in 2010, creating an international framework for global collaboration, data sharing, quality assurance and enhancing accurate annotation of the genome-encoded proteome. During the subsequent decade, the HPP established collaborations, developed guidelines and metrics, and undertook reanalysis of previously deposited community data, continuously increasing the coverage of the human proteome. On the occasion of the HPP’s tenth anniversary, we here report a 90.4% complete high-stringency human proteome blueprint. This knowledge is essential for discerning molecular processes in health and disease, as we demonstrate by highlighting potential roles the human proteome plays in our understanding, diagnosis and treatment of cancers, cardiovascular and infectious diseases.

Near-Direct Bandgap WSe2/ReS2 Type-II pn Heterojunction for Enhanced Ultrafast Photodetection and High-Performance Photovoltaics.

Abstract: Pn heterojunctions comprising layered van der Waals (vdW) semiconductors have been used to demonstrate current-rectifiers, photodetectors, and photovoltaic devices. However, a direct or near-direct...

Transition Metal Catalyzed Enantioselective C(sp2)–H Bond Functionalization

Abstract: Direct catalytic transformation of C–H bonds to new functionalities has provided a powerful strategy to synthesize complex molecular scaffolds in a straightforward way. Unstinting efforts of the sy...

Assessing magnetic and inductive thermal properties of various surfactants functionalised Fe 3 O 4 nanoparticles for hyperthermia

Abstract: This work reports the fabrication of magnetite (Fe3O4) nanoparticles (NPs) coated with various biocompatible surfactants such as glutamic acid (GA), citric acid (CA), polyethylene glycol (PEG), polyvinylpyrrolidine (PVP), ethylene diamine (EDA) and cetyl-trimethyl ammonium bromide (CTAB) via co-precipitation method and their comparative inductive heating ability for hyperthermia (HT) applications. X-ray and electron diffraction analyses validated the formation of well crystallined inverse spinel structured Fe3O4 NPs (crystallite size of ~ 8-10 nm). Magnetic studies confirmed the superparamagnetic (SPM) behaviour for all the NPs with substantial magnetisation (63-68 emu/g) and enhanced magnetic susceptibility is attributed to the greater number of occupations of Fe2+ ions in the lattice as revealed by X-ray photoelectron spectroscopy (XPS). Moreover, distinctive heating response (specific absorption rate, SAR from 130 to 44 W/g) of NPs with similar size and magnetisation is observed. The present study was successful in establishing a direct correlation between relaxation time (~ 9.42-15.92 ns) and heating efficiency of each surface functionalised NPs. Moreover, heat dissipated in different surface grafted NPs is found to be dependent on magnetic susceptibility, magnetic anisotropy and magnetic relaxation time. These results open very promising avenues to design surface functionalised magnetite NPs for effective HT applications.

Examining effects of the COVID-19 national lockdown on ambient air quality across urban India

Abstract: Lockdown seems the most effective way to prevent the spread of Coronavirus disease (COVID-19) as no vaccine is currently available in the market to cure it Thus, India has enforced nation-wide lockdown from 25th March to lower the spread of this contagious virus and associated illness This study aims to quantify the changes in pollution levels as well as meteorology during the 6-weeks COVID-19 lockdown over 17 cities of India for 5 major criteria pollutants using publicly available air quality data Hourly averaged data is accessed from the air quality monitoring stations during the lockdown and immediate pre-lockdown periods and also corresponding data from the previous year (2019) During the lockdown, PM25, PM10, NO2, and CO reduced significantly with relatively small changes in meteorological conditions compared to the pre-lockdown period The highest decline is observed over Ahmedabad (68%), Delhi (71%), Bangalore (87%), and Nagpur (63%) for PM25, PM10, NO2, and CO, respectively The Northern region shows the highest decline for all the pollutants with most days below NAAQS during lockdown—86%, 68%, and 100% compared to 18%, 0%, and 38% in 2019 for PM25, PM10, and NO2, respectively The smaller cities Dewas and Jorapokhar show lesser improvement with only 3% and 16% improvement in days under NAAQS for PM25 SO2 is the least affected pollutant with little improvement The major decline is observed during 7–10 am and 7–10 pm hours of the day for PM25, PM10, NO2, and CO with more than 40% reduction The meteorological changes are very small and heterogeneous over India showing a similar extent of changes compared to the previous year but the pollution levels have reduced significantly Thus, the sharp decline in pollutant concentration during the ~6 weeks period national lockdown can be attributed to the reduced economic and transport activities

Model Comparison from LIGO-Virgo Data on GW170817's Binary Components and Consequences for the Merger Remnant

Abstract: GW170817 is the very first observation of gravitational waves originating from the coalescence of two compact objects in the mass range of neutron stars, accompanied by electromagnetic counterparts, and offers an opportunity to directly probe the internal structure of neutron stars. We perform Bayesian model selection on a wide range of theoretical predictions for the neutron star equation of state. For the binary neutron star hypothesis, we find that we cannot rule out the majority of theoretical models considered. In addition, the gravitational-wave data alone does not rule out the possibility that one or both objects were low-mass black holes. We discuss the possible outcomes in the case of a binary neutron star merger, finding that all scenarios from prompt collapse to long-lived or even stable remnants are possible. For long-lived remnants, we place an upper limit of 1.9 kHz on the rotation rate. If a black hole was formed any time after merger and the coalescing stars were slowly rotating, then the maximum baryonic mass of non-rotating neutron stars is at most 3.05M⊙, and three equations of state considered here can be ruled out. We obtain a tighter limit of 2.67M⊙ for the case that the merger results in a hypermassive neutron star.

COMET Phase-I Technical Design Report

Abstract: The Technical Design for the COMET Phase-I experiment is presented in this paper. COMET is an experiment at J-PARC, Japan, which will search for neutrinoless conversion of muons into electrons in the field of an aluminum nucleus (⁠|$\mu$|–|$e$| conversion, |$\mu^{-}N \rightarrow e^{-}N$|⁠); a lepton flavor-violating process. The experimental sensitivity goal for this process in the Phase-I experiment is |$3.1\times10^{-15}$|⁠, or 90% upper limit of a branching ratio of |$7\times 10^{-15}$|⁠, which is a factor of 100 improvement over the existing limit. The expected number of background events is 0.032. To achieve the target sensitivity and background level, the 3.2 kW 8 GeV proton beam from J-PARC will be used. Two types of detectors, CyDet and StrECAL, will be used for detecting the |$\mu$|–|$e$| conversion events, and for measuring the beam-related background events in view of the Phase-II experiment, respectively. Results from simulation on signal and background estimations are also described.

FusAtNet: Dual Attention based SpectroSpatial Multimodal Fusion Network for Hyperspectral and LiDAR Classification

Abstract: With recent advances in sensing, multimodal data is becoming easily available for various applications, especially in remote sensing (RS), where many data types like multispectral imagery (MSI), hyperspectral imagery (HSI), LiDAR etc. are available. Effective fusion of these multisource datasets is becoming important, for these multimodality features have been shown to generate highly accurate land-cover maps. However, fusion in the context of RS is non-trivial considering the redundancy involved in the data and the large domain differences among multiple modalities. In addition, the feature extraction modules for different modalities hardly interact among themselves, which further limits their semantic relatedness. As a remedy, we propose a feature fusion and extraction framework, namely FusAtNet, for collective land-cover classification of HSIs and LiDAR data in this paper. The proposed framework effectively utilizses HSI modality to generate an attention map using "self-attention" mechanism that highlights its own spectral features. Similarly, a "cross-attention" approach is simultaneously used to harness the LiDAR derived attention map that accentuates the spatial features of HSI. These attentive spectral and spatial representations are then explored further along with the original data to obtain modality-specific feature embeddings. The modality oriented joint spectro-spatial information thus obtained, is subsequently utilized to carry out the land-cover classification task. Experimental evaluations on three HSILiDAR datasets show that the proposed method achieves the state-of-the-art classification performance, including on the largest HSI-LiDAR dataset available, University of Houston (Data Fusion Contest - 2013), opening new avenues in multimodal feature fusion for classification.

Efficient Domain Generalization via Common-Specific Low-Rank Decomposition

Abstract: Domain generalization refers to the task of training a model which generalizes to new domains that are not seen during training. We present CSD (Common Specific Decomposition), for this setting,which jointly learns a common component (which generalizes to new domains) and a domain specific component (which overfits on training domains). The domain specific components are discarded after training and only the common component is retained. The algorithm is extremely simple and involves only modifying the final linear classification layer of any given neural network architecture. We present a principled analysis to understand existing approaches, provide identifiability results of CSD,and study effect of low-rank on domain generalization. We show that CSD either matches or beats state of the art approaches for domain generalization based on domain erasure, domain perturbed data augmentation, and meta-learning. Further diagnostics on rotated MNIST, where domains are interpretable, confirm the hypothesis that CSD successfully disentangles common and domain specific components and hence leads to better domain generalization.