arXiv subjects

Rishabh Jain

Publications and source records attributed to Rishabh Jain.

67 records · Page 4Linked to original sources

arXiv2020

Numerical scheme for the far-out-of-equilibrium time-dependent Boltzmann collision operator: 1D second-degree momentum discretisation and adaptive time stepping

Study of far-from-equilibrium thermalization dynamics in quantum materials, including the dynamics of different types of quasiparticles, is becoming increasingly crucial. However, the inherent complexity of either the full quantum mechanical treatment or the solution of the scattering integral in the Boltzmann approach, has significantly limited the progress in this domain. In our previous work we had developed a solver to calculate the scattering integral in the Boltzmann equation. The solver is free of any approximation (no linearisation of the scattering operator, no close-to-equilibrium approximation, full non-analytic dispersions, full account of Pauli factors, and no limit to low order scattering) \cite{Michael}. Here we extend it to achieve a higher order momentum space convergence by extending to second degree basis functions.cWe further use an adaptive time stepper, achieving a significant improvement in the numerical performance. Moreover we show adaptive time stepping can prevent intrinsic instabilities in the time propagation of the Boltzmann scattering operator. This work makes the numerical time propagation of the full Boltzmann scattering operator efficient, stable and minimally reliant on human supervision.

physics.comp-ph↗

arXiv2020

Dialog without Dialog Data: Learning Visual Dialog Agents from VQA Data

Can we develop visually grounded dialog agents that can efficiently adapt to new tasks without forgetting how to talk to people? Such agents could leverage a larger variety of existing data to generalize to new tasks, minimizing expensive data collection and annotation. In this work, we study a setting we call "Dialog without Dialog", which requires agents to develop visually grounded dialog models that can adapt to new tasks without language level supervision. By factorizing intention and language, our model minimizes linguistic drift after fine-tuning for new tasks. We present qualitative results, automated metrics, and human studies that all show our model can adapt to new tasks and maintain language quality. Baselines either fail to perform well at new tasks or experience language drift, becoming unintelligible to humans. Code has been made available at https://github.com/mcogswell/dialog_without_dialog

cs.CV↗

arXiv2019

nocaps: novel object captioning at scale

Image captioning models have achieved impressive results on datasets containing limited visual concepts and large amounts of paired image-caption training data. However, if these models are to ever function in the wild, a much larger variety of visual concepts must be learned, ideally from less supervision. To encourage the development of image captioning models that can learn visual concepts from alternative data sources, such as object detection datasets, we present the first large-scale benchmark for this task. Dubbed 'nocaps', for novel object captioning at scale, our benchmark consists of 166,100 human-generated captions describing 15,100 images from the OpenImages validation and test sets. The associated training data consists of COCO image-caption pairs, plus OpenImages image-level labels and object bounding boxes. Since OpenImages contains many more classes than COCO, nearly 400 object classes seen in test images have no or very few associated training captions (hence, nocaps). We extend existing novel object captioning models to establish strong baselines for this benchmark and provide analysis to guide future work on this task.

cs.CV↗

arXiv2019

On Model Stability as a Function of Random Seed

In this paper, we focus on quantifying model stability as a function of random seed by investigating the effects of the induced randomness on model performance and the robustness of the model in general. We specifically perform a controlled study on the effect of random seeds on the behaviour of attention, gradient-based and surrogate model based (LIME) interpretations. Our analysis suggests that random seeds can adversely affect the consistency of models resulting in counterfactual interpretations. We propose a technique called Aggressive Stochastic Weight Averaging (ASWA)and an extension called Norm-filtered Aggressive Stochastic Weight Averaging (NASWA) which improves the stability of models over random seeds. With our ASWA and NASWA based optimization, we are able to improve the robustness of the original model, on average reducing the standard deviation of the model's performance by 72%.

cs.LG↗

arXiv2019

Flavor Changing Heavy Higgs Interactions with Leptons at Hadron Colliders

In a general two Higgs doublet model, we study flavor changing neutral Higgs (FCNH) decays into leptons at hadron colliders, $pp \to ϕ^0 \to τ^\mpμ^\pm +X$, where $ϕ^0$ could be a CP-even scalar ($h^0$, $H^0$) or a CP-odd pseudoscalar ($A^0$). The light Higgs boson $h^0$ is found to resemble closely the Standard Model Higgs boson at the Large Hadron Collider. In the alignment limit of $\cos(β-α) \cong 0$ for $h^0$--$H^0$ mixing, FCNH couplings of $h^0$ are naturally suppressed, but such couplings of the heavier $H^0, A^0$ are sustained by $\sin(β-α) \simeq 1$. We evaluate physics backgrounds from dominant processes with realistic acceptance cuts and tagging efficiencies. We find promising results for $\sqrt{s} = 14$ TeV, which we extend further to $\sqrt{s} = 27$ TeV and 100 TeV future pp colliders.

hep-ph↗

arXiv2019

Model Explanations under Calibration

Explaining and interpreting the decisions of recommender systems are becoming extremely relevant both, for improving predictive performance, and providing valid explanations to users. While most of the recent interest has focused on providing local explanations, there has been a much lower emphasis on studying the effects of model dynamics and its impact on explanation. In this paper, we perform a focused study on the impact of model interpretability in the context of calibration. Specifically, we address the challenges of both over-confident and under-confident predictions with interpretability using attention distribution. Our results indicate that the means of using attention distributions for interpretability are highly unstable for un-calibrated models. Our empirical analysis on the stability of attention distribution raises questions on the utility of attention for explainability.

cs.IR↗

arXiv2019

Charming Top Decays with Flavor Changing Neutral Higgs Boson and WW at Hadron Colliders

We investigate the prospects for discovering a top quark decaying into one light Higgs boson ($h^0$) along with a charm quark ($c$) in top quark pair production at the CERN Large Hadron Collider (LHC) and future hadron colliers. A general two Higgs doublet model is adopted to study the signature of flavor changing neutral Higgs (FCNH) interactions with $t \to c h^0$, followed by $h^0 \to WW^* \to \ell^+ \ell^- +\notE_T$, where $h^0$ is the CP-even Higgs boson and $\notE_T$ stands for missing transverse energy from neutrinos. We study the discovery potential for this FCNH signal and physics background from dominant processes with realistic acceptance cuts as well as tagging and mistagging efficiencies. Promising results are found for the LHC running at 13 TeV and 14 TeV center-of-mass energy as well as future pp colliders at 27 TeV and 100 TeV.

hep-ph↗

arXiv2019

EvalAI: Towards Better Evaluation Systems for AI Agents

We introduce EvalAI, an open source platform for evaluating and comparing machine learning (ML) and artificial intelligence algorithms (AI) at scale. EvalAI is built to provide a scalable solution to the research community to fulfill the critical need of evaluating machine learning models and agents acting in an environment against annotations or with a human-in-the-loop. This will help researchers, students, and data scientists to create, collaborate, and participate in AI challenges organized around the globe. By simplifying and standardizing the process of benchmarking these models, EvalAI seeks to lower the barrier to entry for participating in the global scientific effort to push the frontiers of machine learning and artificial intelligence, thereby increasing the rate of measurable progress in this domain.

cs.AI↗

arXiv2016

The Jeans theorem and the "Tolman-Oppenheimer-Volkov equation" in an exact wave solution of R^2 gravity

Corda, Mosquera Cuesta and Lorduy Gomez have shown that spherically symmetric stationary states can be used as a model for galaxies in the framework of the linearized R^2 gravity. Those states could represent a partial solution to the Dark Matter Problem. Here we discuss an improvement of this work. In fact, as the star density is a functional of the invariants of the associated Vlasov equation, we show that any of these invariants is in its turn a functional of the local energy and the angular momentum. As a consequence, the star density depends only on these two integrals of the Vlasov system. This result is known as the "Jeans theorem". In addition, we find an analogous of the historical Tolman- Oppenheimer-Volkov equation for the system considered in this paper. For the sake of completeness, in the final Section of the paper we consider two additional models which argue that Dark Matter could not be an essential element.

gr-qc↗

arXiv2015

Effect of shear strain on band structure and electronic properties of phosphorene

We present an ab-initio investigation of effects of shear strain on band structure and electronic properties of 2D phosphorene. We carried out DFT calculations to determine the shear stress as a function of shear strain and found the monolayer phosphorene has ultimate strength at shear strain 30% and 35% in armchair and zigzag directions, respectively, and it was also found that the monolayer extends in z direction on applying shear strain in both directions. Additionally, we derived band structures of phosphorene along both directions under shear strain and have shown that band gap in phosphorene decreases along both directions and that phosphorene shows a semi-metal nature on applying shear strain of magnitude 30% in both directions. The electrical conductivity of phosphorene was estimated by effective mass along zigzag and armchair directions and it is shown that the electrical conductivity is far higher along armchair direction, and that with increasing shear strain conductivity increases along armchair, up to ultimate strength, and zigzag directions.

cond-mat.mtrl-sci↗

arXiv2014

Optimizing the For loop: Comparison of For loop and micro For loop

Looping is one of the fundamental logical instructions used for repeating a block of code. It is used in programs across all programming languages. Traditionally, in languages like C, the for loop is used extensively for repeated execution of a block of code, due to its ease for use and simplified representation. This paper proposes a new way of representing the for loop to improve its runtime efficiency and compares the experimental statistics with the traditional for loop representation. It is found that for small number of iterations, the difference in computational time may not be considerable. But given any large number of iterations, the difference is noticeable.

cs.PL↗

arXiv2014

Clear, Concise and Effective UI: Opinion and Suggestions

The most important aspect of any Software is the operability for the intended audience. This factor of operability is encompassed in the user interface, which serves as the only window to the features of the system. It is thus essential that the User Interface provided is robust, concise and lucid. Presently there are no properly defined rules or guidelines for user interface design enabling a perfect design, since such a system cannot be perceived. This article aims at providing suggestions in the design of the User Interface, which would make it easier for the user to navigate through the system features and also the developers to guide the users towards better utilization of the features.

cs.HC↗

arXiv2014

Mining and Analyzing Twitter trends: Frequency based ranking of descriptive Tweets

One of the major sources of trending news, events and opinion in the current age is micro blogging. Twitter, being one of them, is extensively used to mine data about public responses and event updates. This paper intends to propose methods to filter tweets to obtain the most accurately descriptive tweets, which communicates the content of the trend. It also potentially ranks the tweets according to relevance. The principle behind the ranking mechanism would be the assumed tendencies in the natural language used by the users. The mapping frequencies of occurrence of words and related hash tags is used to create a weighted score for each tweet in the sample space obtained from twitter on a particular trend.

cs.SI↗