arXiv subjects

Supriyo Ghosh

Publications and source records attributed to Supriyo Ghosh.

At least 37 records · Page 2Linked to original sources

arXiv2024

Streetwise Agents: Empowering Offline RL Policies to Outsmart Exogenous Stochastic Disturbances in RTC

The difficulty of exploring and training online on real production systems limits the scope of real-time online data/feedback-driven decision making. The most feasible approach is to adopt offline reinforcement learning from limited trajectory samples. However, after deployment, such policies fail due to exogenous factors that temporarily or permanently disturb/alter the transition distribution of the assumed decision process structure induced by offline samples. This results in critical policy failures and generalization errors in sensitive domains like Real-Time Communication (RTC). We solve this crucial problem of identifying robust actions in presence of domain shifts due to unseen exogenous stochastic factors in the wild. As it is impossible to learn generalized offline policies within the support of offline data that are robust to these unseen exogenous disturbances, we propose a novel post-deployment shaping of policies (Streetwise), conditioned on real-time characterization of out-of-distribution sub-spaces. This leads to robust actions in bandwidth estimation (BWE) of network bottlenecks in RTC and in standard benchmarks. Our extensive experimental results on BWE and other standard offline RL benchmark environments demonstrate a significant improvement ($\approx$ 18% on some scenarios) in final returns wrt. end-user metrics over state-of-the-art baselines.

cs.LG↗

arXiv2024

Exploring large language models for microstructure evolution in materials

There is a significant potential for coding skills to transition fully to natural language in the future. In this context, large language models (LLMs) have shown impressive natural language processing abilities to generate sophisticated computer code for research tasks in various domains. We report the first study on the applicability of LLMs to perform computer experiments on microstructure pattern formation in model materials. In particular, we exploit LLM's ability to generate code for solving various types of phase-field-based partial differential equations (PDEs) that integrate additional physics to model material microstructures. The results indicate that LLMs have a remarkable capacity to generate multi-physics code and can effectively deal with materials microstructure problems up to a certain complexity. However, for complex multi-physics coupled PDEs for which a detailed understanding of the problem is required, LLMs fail to perform the task efficiently, since much more detailed instructions with many iterations of the same query are required to generate the desired output. Nonetheless, at their current stage of development and potential future advancements, LLMs offer a promising outlook for accelerating materials education and research by supporting beginners and experts in their physics-based methodology. We hope this paper will spur further interest to leverage LLMs as a supporting tool in the integrated computational materials engineering (ICME) approach to materials modeling and design.

cond-mat.mtrl-sci↗

arXiv2024

Kinetics of orbital ordering in cooperative Jahn-Teller models: Machine-learning enabled large-scale simulations

We present a scalable machine learning (ML) force-field model for the adiabatic dynamics of cooperative Jahn-Teller (JT) systems. Large scale dynamical simulations of the JT model also shed light on the orbital ordering dynamics in colossal magnetoresistance manganites. The JT effect in these materials describes the distortion of local oxygen octahedra driven by a coupling to the orbital degrees of freedom of $e_g$ electrons. An effective electron-mediated interaction between the local JT modes leads to a structural transition and the emergence of long-range orbital order at low temperatures. Assuming the principle of locality, a deep-learning neural-network model is developed to accurately and efficiently predict the electron-induced forces that drive the dynamical evolution of JT phonons. A group-theoretical method is utilized to develop a descriptor that incorporates the combined orbital and lattice symmetry into the ML model. Large-scale Langevin dynamics simulations, enabled by the ML force-field models, are performed to investigate the coarsening dynamics of the composite JT distortion and orbital order after a thermal quench. The late-stage coarsening of orbital domains exhibits pronounced freezing behaviors which are likely related to the unusual morphology of the domain structures. Our work highlights a promising avenue for multi-scale dynamical modeling of correlated electron systems.

cond-mat.str-el↗

arXiv2024

X-lifecycle Learning for Cloud Incident Management using LLMs

Incident management for large cloud services is a complex and tedious process and requires significant amount of manual efforts from on-call engineers (OCEs). OCEs typically leverage data from different stages of the software development lifecycle [SDLC] (e.g., codes, configuration, monitor data, service properties, service dependencies, trouble-shooting documents, etc.) to generate insights for detection, root causing and mitigating of incidents. Recent advancements in large language models [LLMs] (e.g., ChatGPT, GPT-4, Gemini) created opportunities to automatically generate contextual recommendations to the OCEs assisting them to quickly identify and mitigate critical issues. However, existing research typically takes a silo-ed view for solving a certain task in incident management by leveraging data from a single stage of SDLC. In this paper, we demonstrate that augmenting additional contextual data from different stages of SDLC improves the performance of two critically important and practically challenging tasks: (1) automatically generating root cause recommendations for dependency failure related incidents, and (2) identifying ontology of service monitors used for automatically detecting incidents. By leveraging 353 incident and 260 monitor dataset from Microsoft, we demonstrate that augmenting contextual information from different stages of the SDLC improves the performance over State-of-The-Art methods.

cs.NI↗

arXiv2024

Dependency Aware Incident Linking in Large Cloud Systems

Despite significant reliability efforts, large-scale cloud services inevitably experience production incidents that can significantly impact service availability and customer's satisfaction. Worse, in many cases one incident can lead to multiple downstream failures due to cascading effects that creates several related incidents across different dependent services. Often time On-call Engineers (OCEs) examine these incidents in silos that lead to significant amount of manual toil and increase the overall time-to-mitigate incidents. Therefore, developing efficient incident linking models is of paramount importance for grouping related incidents into clusters so as to quickly resolve major outages and reduce on-call fatigue. Existing incident linking methods mostly leverages textual and contextual information of incidents (e.g., title, description, severity, impacted components), thus failing to leverage the inter-dependencies between services. In this paper, we propose the dependency-aware incident linking (DiLink) framework which leverages both textual and service dependency graph information to improve the accuracy and coverage of incident links not only coming from same service, but also from different services and workloads. Furthermore, we propose a novel method to align the embeddings of multi-modal (i.e., textual and graphical) data using Orthogonal Procrustes. Extensive experimental results on real-world incidents from 5 workloads of Microsoft demonstrate that our alignment method has an F1-score of 0.96 (14% gain over current state-of-the-art methods). We are also in the process of deploying this solution across 610 services from these 5 workloads for continuously supporting OCEs improving incident management and reducing manual toil.

cs.DC↗

arXiv2024

Automated Root Causing of Cloud Incidents using In-Context Learning with GPT-4

Root Cause Analysis (RCA) plays a pivotal role in the incident diagnosis process for cloud services, requiring on-call engineers to identify the primary issues and implement corrective actions to prevent future recurrences. Improving the incident RCA process is vital for minimizing service downtime, customer impact and manual toil. Recent advances in artificial intelligence have introduced state-of-the-art Large Language Models (LLMs) like GPT-4, which have proven effective in tackling various AIOps problems, ranging from code authoring to incident management. Nonetheless, the GPT-4 model's immense size presents challenges when trying to fine-tune it on user data because of the significant GPU resource demand and the necessity for continuous model fine-tuning with the emergence of new data. To address the high cost of fine-tuning LLM, we propose an in-context learning approach for automated root causing, which eliminates the need for fine-tuning. We conduct extensive study over 100,000 production incidents, comparing several large language models using multiple metrics. The results reveal that our in-context learning approach outperforms the previous fine-tuned large language models such as GPT-3 by an average of 24.8\% across all metrics, with an impressive 49.7\% improvement over the zero-shot model. Moreover, human evaluation involving actual incident owners demonstrates its superiority over the fine-tuned model, achieving a 43.5\% improvement in correctness and an 8.7\% enhancement in readability. The impressive results demonstrate the viability of utilizing a vanilla GPT model for the RCA task, thereby avoiding the high computational and maintenance costs associated with a fine-tuned model.

cs.CL↗

arXiv2023

Mitigating Modal Noise in Multimode Circular Fibres by Optical Agitation using a Galvanometer

Modal noise appears due to the non-uniform and unstable distribution of light intensity among the finite number of modes in multimode fibres. It is an important limiting factor in measuring radial velocity precisely by fibre-fed high-resolution spectrographs. The problem can become particularly severe as the fibre's core become smaller and the number of modes that can propagate reduces. Thus, mitigating modal noise in relatively small core fibres still remains a challenge. We present here a novel technique to suppress modal noise. Two movable mirrors in the form of a galvanometer reimage the mode-pattern of an input fibre to an output fibre. The mixing of modes coupled to the output fibre can be controlled by the movement of mirrors applying two sinusoidal signals through a voltage generator. We test the technique for four multimode circular fibres: 10 and 50 micron step-index, 50 micron graded-index, and a combination of 50 micron graded-index and 5:1 tapered fibres (GI50t). We present the results of mode suppression both in terms of the direct image of the output fibre and spectrum of white light obtained with the high-resolution spectrograph. We found that the galvanometer mitigated modal noise in all the tested fibres, but was most useful for smaller core fibres. However, there is a trade-off between the modal noise reduction and light-loss. The GI50t provides the best result with about 60% mitigation of modal noise at a cost of about 5% output light-loss. Our solution is easy to use and can be implemented in fibre-fed spectrographs.

astro-ph.IM↗

arXiv2023

Automatic Root Cause Analysis via Large Language Models for Cloud Incidents

Ensuring the reliability and availability of cloud services necessitates efficient root cause analysis (RCA) for cloud incidents. Traditional RCA methods, which rely on manual investigations of data sources such as logs and traces, are often laborious, error-prone, and challenging for on-call engineers. In this paper, we introduce RCACopilot, an innovative on-call system empowered by the large language model for automating RCA of cloud incidents. RCACopilot matches incoming incidents to corresponding incident handlers based on their alert types, aggregates the critical runtime diagnostic information, predicts the incident's root cause category, and provides an explanatory narrative. We evaluate RCACopilot using a real-world dataset consisting of a year's worth of incidents from Microsoft. Our evaluation demonstrates that RCACopilot achieves RCA accuracy up to 0.766. Furthermore, the diagnostic information collection component of RCACopilot has been successfully in use at Microsoft for over four years.

cs.SE↗

arXiv2023

Quantum Integrability and Chaos in periodic Toda Lattice with Balanced Loss-Gain

We consider equal-mass quantum Toda lattice with balanced loss-gain for two and three particles. The two-particle Toda lattice is integrable and two integrals of motion which are in involution have been found. The bound-state energy and the corresponding eigenfunctions have been obtained numerically for a few low-lying states. The three-particle quantum Toda lattice with balanced loss-gain and velocity mediated coupling admits mixed phases of integrability and chaos depending on the value of the loss-gain parameter. We have obtained analytic expressions for two integrals of motion which are in involution. Although an analytic expression for the third integral has not been found, the numerical investigation suggests integrability below a critical value of the loss-gain strength and chaos above this critical value. The level spacing distribution changes from the Wigner-Dyson to the Poisson distribution as the loss-gain parameter passes through this critical value and approaches zero. An identical behaviour is seen in terms of the gap-ratio distribution of the energy levels. The existence of mixed phases of quantum integrability and chaos in the specified ranges of the loss-gain parameter has also been confirmed independently via the study of level repulsion and complexity in higher order excited states.

nlin.CD↗

arXiv2023

Solvable Limits of a class of generalized Vector Nonlocal Nonlinear Schrödinger equation with balanced loss-gain

We consider a class of one dimensional Vector Nonlocal Non-linear Schrödinger Equation (VNNLSE) in an external complex potential with time-modulated Balanced Loss-Gain(BLG) and Linear Coupling(LC) among the components of Schrödinger fields, and space-time dependent nonlinear strength. The system admits Lagrangian and Hamiltonian formulations under certain conditions. It is shown that various dynamical variables like total power, $\cal{PT}$-symmetric Hamiltonian, width of the wave-packet and its speed of growth, etc. are real-valued despite the Hamiltonian density being complex-valued. We study the exact solvability of the generic VNNLSE with or without a Hamiltonian formulation. In the first part, we study time-evolution of moments which are analogous to space-integrals of Stokes variables and find condition for existence of solutions which are bounded in time. In the second part, we use a non-unitary transformation followed by a coordinate transformation to map the VNNLSE to various solvable equations. The cordinate transformation is not required at all for the limiting case when non-unitary transformation reduces to pseudo-unitary transformation. The exact solutions are bounded in time for the same condition which is obtained through the study of time-evolution of moments. Various exact solutions of the VNNLSE are presented.

nlin.SI↗

arXiv2023

Theoretical investigations on Kerr and Faraday rotations in topological multi-Weyl Semimetals

Motivated by the recent proposal of giant Kerr rotation in WSMs, we investigate the Kerr and Faraday rotations in time-reversal broken multi-Weyl semimetals (mWSMs) in the absence of an external magnetic field. Using the framework of Kubo response theory, we find that both the longitudinal and transverse components of the optical conductivity in mWSMs are modified by the topological charge ($n$). Engendered by the optical Hall conductivity, we show in the thin film limit that, while the giant Kerr rotation and corresponding ellipticity are independent of $n$, the Faraday rotation and its ellipticity angle scale as $n$ and $n^2$, respectively. In contrast, the polarization rotation in semi-infinite mWSMs is dominated by the axion field showing $n$ dependence. In particular, the magnitude of Kerr (Faraday) angle decreases (increases) with increasing $n$ in Faraday geometry, whereas in Voigt geometry, it depicts different $n$-dependencies in different frequency regimes. The obtained results on the behavior of polarization rotations in mWSMs could be used in experiments as a probe to distinguish single, double, and triple WSMs, as well as discriminate the surfaces of mWSMs with and without hosting Fermi arcs.

cond-mat.mes-hall↗

arXiv2023

Deep Offline Reinforcement Learning for Real-world Treatment Optimization Applications

There is increasing interest in data-driven approaches for recommending optimal treatment strategies in many chronic disease management and critical care applications. Reinforcement learning methods are well-suited to this sequential decision-making problem, but must be trained and evaluated exclusively on retrospective medical record datasets as direct online exploration is unsafe and infeasible. Despite this requirement, the vast majority of treatment optimization studies use off-policy RL methods (e.g., Double Deep Q Networks (DDQN) or its variants) that are known to perform poorly in purely offline settings. Recent advances in offline RL, such as Conservative Q-Learning (CQL), offer a suitable alternative. But there remain challenges in adapting these approaches to real-world applications where suboptimal examples dominate the retrospective dataset and strict safety constraints need to be satisfied. In this work, we introduce a practical and theoretically grounded transition sampling approach to address action imbalance during offline RL training. We perform extensive experiments on two real-world tasks for diabetes and sepsis treatment optimization to compare performance of the proposed approach against prominent off-policy and offline RL baselines (DDQN and CQL). Across a range of principled and clinically relevant metrics, we show that our proposed approach enables substantial improvements in expected health outcomes and in accordance with relevant practice and safety guidelines.

cs.LG↗

arXiv2023

Non-linear Schr$\ddot{o}$dinger equation with time-dependent balanced loss-gain and space-time modulated non-linear interaction

We consider a class of one dimensional vector Non-linear Schr$\ddot{o}$dinger Equation(NLSE) in an external complex potential with Balanced Loss-Gain(BLG) and Linear Coupling(LC) among the components of the Schr$\ddot{o}$dinger field. The solvability of the generic system is investigated for various combinations of time modulated LC and BLG terms, space-time dependent strength of the nonlinear interaction and complex potential. We use a non-unitary transformation followed by a reformulation of the differential equation in a new coordinate system to map the NLSE to solvable equations. Several physically motivated examples of exactly solvable systems are presented for various combinations of LC and BLG, external complex potential and nonlinear interaction. Exact localized nonlinear modes with spatially constant phase may be obtained for any real potential for which the corresponding linear Schr$\ddot{o}$dinger equation is solvable. A method based on supersymmetric quantum mechanics is devised to construct exact localized nonlinear modes for a class of complex potentials. The real superpotential corresponding to any exactly solved linear Schr$\ddot{o}$dinger equation may be used to find a complex-potential for which exact localized nonlinear modes for the NLSE can be obtained. The solutions with singular phases are obtained for a few complex potentials.

math-ph↗

arXiv2023

Quiescence of an Outburst of a Low-Mass Young Stellar Object: LDN1415-IRS

LDN1415-IRS, a low-mass young stellar object (YSO) went into an outburst between 2001 and 2006, illuminating a surrounding nebula, LDN1415-Neb. LDN1415-Neb was found to have brightened by I=3.77 mag by April 2006. The optical light curve covering $\sim$ 15.5 years, starting from October 2006 to January 2022, is presented in this study. The initial optical spectrum indicated the presence of winds in the system but the subsequent spectra have no wind indicators. The declining light curve and the absence of the P-Cygni profile in later epoch spectra indicate that the star and nebula system is retrieving back from its outburst state. Two Herbig-Haro objects (HHOs) are positioned linearly with respect to the optical brightness peak of the nebula, probably indicating the circumstellar disk being viewed edge-on. Our recent deep near-infrared (NIR) imaging using TANSPEC has revealed the presence of a nearby star-like source, south of the LDN1415-IRS, at an angular distance of $\sim$ 5.4 arcsec.

astro-ph.SR↗

arXiv2023

Recommending Root-Cause and Mitigation Steps for Cloud Incidents using Large Language Models

Incident management for cloud services is a complex process involving several steps and has a huge impact on both service health and developer productivity. On-call engineers require significant amount of domain knowledge and manual effort for root causing and mitigation of production incidents. Recent advances in artificial intelligence has resulted in state-of-the-art large language models like GPT-3.x (both GPT-3.0 and GPT-3.5), which have been used to solve a variety of problems ranging from question answering to text summarization. In this work, we do the first large-scale study to evaluate the effectiveness of these models for helping engineers root cause and mitigate production incidents. We do a rigorous study at Microsoft, on more than 40,000 incidents and compare several large language models in zero-shot, fine-tuned and multi-task setting using semantic and lexical metrics. Lastly, our human evaluation with actual incident owners show the efficacy and future potential of using artificial intelligence for resolving cloud incidents.

cs.SE↗

arXiv2023

Spectroscopy of 9 eruptive young variables using TANSPEC

In recent times, 3.6m Devasthal Optical Telescope (DOT) has installed an optical to near infra-red spectrograph, TANSPEC, which provides spectral coverage from 0.55-2.5 microns. Using TANSPEC, we have obtained a single epoch spectrum of a set 9 FUors and EXors. We have analysed line profiles of the sources and compared with the previously published spectra of these objects. Comparing the line profile shapes with the existing theoretical predictions, we have tried to interpret the physical processes that are responsible for the current disc evolution and the present accretion dynamics. Our study has shown the importance for time evolved spectroscopic studies for better understanding the evolution of the accretion mechanisms. This in turn can help in better categorisation of the young stars displaying episodic accretion behaviour.

astro-ph.SR↗

arXiv2022

pyTANSPEC: A Data Reduction Package for TANSPEC

The TIFR-ARIES Near Infrared Spectrometer (TANSPEC) instrument provides simultaneous wavelength coverage from 0.55 to 2.5 micron, mounted on India's largest ground-based telescope, 3.6-m Devasthal Optical Telescope at Nainital, India. The TANSPEC offers three modes of observations, imaging with various filters, spectroscopy in the low-resolution prism mode with derived R~ 100-400 and the high-resolution cross-dispersed mode (XD-mode) with derived median R~ 2750 for a slit of width 0.5 arcsec. In the XD-mode, ten cross-dispersed orders are packed in the 2048 x 2048 pixels detector to cover the full wavelength regime. As the XD-mode is most utilized as well as for consistent data reduction for all orders and to reduce data reduction time, a dedicated pipeline is at the need. In this paper, we present the code for the TANSPEC XD-mode data reduction, its workflow, input/output files, and a showcase of its implementation on a particular dataset. This publicly available pipeline pyTANSPEC is fully developed in Python and includes nominal human intervention only for the quality assurance of the reduced data. Two customized configuration files are used to guide the data reduction. The pipeline creates a log file for all the fits files in a given data directory from its header, identifies correct frames (science, continuum and calibration lamps) based on the user input, and offers an option to the user for eyeballing and accepting/removing of the frames, does the cleaning of raw science frames and yields final wavelength calibrated spectra of all orders simultaneously.

astro-ph.IM↗

arXiv2022

TANSPEC: TIFR-ARIES Near Infrared Spectrometer

We present the design and performance of the TANSPEC, a medium-resolution $0.55-2.5~μ$m cryogenic spectrometer and imager, now in operation at the 3.6-m Devasthal Optical Telescope (DOT), Nainital, India. The TANSPEC provides three modes of operation which include, photometry with broad- and narrow-band filters, spectroscopy with short slits of 20$^{\prime \prime}$ length and different widths (from 0.5$^{\prime \prime}$ to 4.0$^{\prime \prime}$) in cross-dispersed mode at a resolving power R of $\sim$2750, and spectroscopy with long slits of 60$^{\prime \prime}$ length and different widths (from 0.5$^{\prime \prime}$ to 4.0$^{\prime \prime}$) in prism mode at a resolving power R of $\sim$100-350. TANSPEC's imager mode provides a field of view of 60$^{\prime \prime} \times 60^{\prime \prime}$ with a plate scale of 0.245$^{\prime \prime}$/pixel on the 3.6-m DOT. The TANSPEC was successfully commissioned during April-May 2019 and the subsequent characterization and astronomical observations are presented here. The TANSPEC has been made available to the worldwide astronomical community for science observations from October 2020.

astro-ph.IM↗