arXiv subjects

Dhruv Sharma

Publications and source records attributed to Dhruv Sharma.

18 recordsLinked to original sources

arXiv2026

AlgoEvolve: LLM-driven Meta-evolution of Algorithmic Trading Programs

Recent work shows that Large Language Models (LLMs) can act as semantic mutation operators for the evolutionary discovery of programs and proofs. Most current applications focus on static coding benchmarks. We extend this paradigm to algorithmic trading. This domain is uniquely challenging because it is noisy, non-stationary, and highly discontinuous. We present AlgoEvolve, an LLM-driven evolutionary framework that generates, evaluates, and iteratively improves executable trading strategies. These strategies are expressed as Python code and evaluated through a rigorous testing protocol. Across multiple experiments, the system exhibits emergent regime-adaptive strategy logic, including autonomous shifts in trading rules. We further introduce a meta-evolutionary outer loop that evolves the prompts guiding program synthesis in the inner loop. This outer loop discovers improved search heuristics. These heuristics balance exploration and exploitation while reducing zero-trade failures. They consistently outperform initial human-designed instructions. The results demonstrate that LLM-based semantic evolution provides a viable approach for continual program synthesis in complex environments.

cs.AI↗

arXiv2025

Phenomenological Ehrenfest Dynamics with Topological and Geometric Phase Effects and the curious case of Elliptical intersection

We present a comprehensive computational framework for simulating nonadiabatic molecular dynamics with explicit inclusion of geometric phase (GP) effects. Our approach is based on a generalized two-level Hamiltonian model that can represent various electronic state crossings - conical intersections, avoided crossings, and elliptic intersections - through appropriate parameterization. We introduce a novel prelooping trajectory initialization scheme, allowing us to encode the memory as an initial phase accumulated due to the adiabatic evolution over the potential energy surface. This is a unified framework to handle different types of level crossings by incorporating Berry curvature-based force corrections to Ehrenfest dynamics, ensuring accurate representation of topological effects. For conical intersections, our method incorporates the theoretically expected phase pi, while for elliptic intersections, it yields a parametrically tunable but loop radius (energy) independent phase different from pi. We also include an eccentricity parameter (e) in the diabatic coupling to model more realistic molecular systems. Numerical simulations demonstrate the consistency of our approach with theoretical predictions for mixing of states and inhibition from mixing due to geometric phase effects. This framework provides a valuable tool for studying quantum-classical interactions in molecular systems where geometric phase effects play a significant role. The elliptical intersection and geometric phase effect opens avenue for the design and discovery of degenerate materials. It produces a fresh look to help develop a new kind of spectroscopy and potential qubit applications. This simple Hamiltonian reveals a pathological phase protection effect E = kr, where k is real, that has great utility in a new spectroscopy design.

cond-mat.mes-hall↗

arXiv2025

Exploring Network-Knowledge Graph Duality: A Case Study in Agentic Supply Chain Risk Analysis

Large Language Models (LLMs) struggle with the complex, multi-modal, and network-native data underlying financial risk. Standard Retrieval-Augmented Generation (RAG) oversimplifies relationships, while specialist models are costly and static. We address this gap with an LLM-centric agent framework for supply chain risk analysis. Our core contribution is to exploit the inherent duality between networks and knowledge graphs (KG). We treat the supply chain network as a KG, allowing us to use structural network science principles for retrieval. A graph traverser, guided by network centrality scores, efficiently extracts the most economically salient risk paths. An agentic architecture orchestrates this graph retrieval alongside data from numerical factor tables and news streams. Crucially, it employs novel ``context shells'' -- descriptive templates that embed raw figures in natural language -- to make quantitative data fully intelligible to the LLM. This lightweight approach enables the model to generate concise, explainable, and context-rich risk narratives in real-time without costly fine-tuning or a dedicated graph database.

cs.AI↗

arXiv2025

Dynamic Shock Recovery in IO Networks with Priority Constraints

Physical risks, such as droughts, floods, rising temperatures, earthquakes, infrastructure failures, and geopolitical conflicts, can ripple through global supply chains, raising costs, and constraining production across industries. Assessing these risks requires understanding not only their immediate effects, but also their cascading impacts. For example, a localized drought can disrupt the supply of critical raw materials such as cobalt or copper, affecting battery and electric vehicle production. Similarly, regional conflicts can impede cross-border trade, leading to broader economic consequences. Building on an existing model of simultaneous supply and demand shocks, we introduce a new propagation algorithm, Priority with Constraint, which modifies standard priority-based rationing by incorporating a minimum supply guarantee for all customers, regardless of their size or priority ranking. We also identify a buffer effect inherent in the Industry Proportional algorithm, which reflects real-world economic resilience. Finally, we extend the static shock propagation model to incorporate dynamic processes. We introduce mechanisms for gradual shock propagation, reflecting demand stickiness and the potential buffering role of inventories, and gradual recovery, modeling the simultaneous recovery of supply capacity and the inherent tendency for demand to return to pre-shock levels. Simulations demonstrate how the interplay between demand adjustment speed and supply recovery speed significantly influences the severity and duration of the economic impact after a shock.

physics.soc-ph↗

arXiv2024

UnMA-CapSumT: Unified and Multi-Head Attention-driven Caption Summarization Transformer

Image captioning is the generation of natural language descriptions of images which have increased immense popularity in the recent past. With this different deep-learning techniques are devised for the development of factual and stylized image captioning models. Previous models focused more on the generation of factual and stylized captions separately providing more than one caption for a single image. The descriptions generated from these suffer from out-of-vocabulary and repetition issues. To the best of our knowledge, no such work exists that provided a description that integrates different captioning methods to describe the contents of an image with factual and stylized (romantic and humorous) elements. To overcome these limitations, this paper presents a novel Unified Attention and Multi-Head Attention-driven Caption Summarization Transformer (UnMA-CapSumT) based Captioning Framework. It utilizes both factual captions and stylized captions generated by the Modified Adaptive Attention-based factual image captioning model (MAA-FIC) and Style Factored Bi-LSTM with attention (SF-Bi-ALSTM) driven stylized image captioning model respectively. SF-Bi-ALSTM-based stylized IC model generates two prominent styles of expression- {romance, and humor}. The proposed summarizer UnMHA-ST combines both factual and stylized descriptions of an input image to generate styled rich coherent summarized captions. The proposed UnMHA-ST transformer learns and summarizes different linguistic styles efficiently by incorporating proposed word embedding fastText with Attention Word Embedding (fTA-WE) and pointer-generator network with coverage mechanism concept to solve the out-of-vocabulary issues and repetition problem. Extensive experiments are conducted on Flickr8K and a subset of FlickrStyle10K with supporting ablation studies to prove the efficiency and efficacy of the proposed framework.

cs.CV↗

arXiv2024

Multi-Class Abnormality Classification Task in Video Capsule Endoscopy

In this work for Capsule Vision Challenge 2024, we addressed the challenge of multiclass anomaly classification in video capsule Endoscopy (VCE)[1] with a variety of deep learning models, ranging from custom CNNs to advanced transformer architectures. The purpose is to correctly classify diverse gastrointestinal disorders, which is critical for increasing diagnostic efficiency in clinical settings. We started with a baseline CNN model and improved performance with ResNet[2] for better feature extraction, followed by Vision Transformer (ViT)[3] to capture global dependencies. We further improve the results by using Multiscale Vision Transformer (MViT)[4] for improved hierarchical feature extraction, while Dual Attention Vision Transformer (DaViT) [5] delivered best results by combining spatial and channel attention methods. Our best balanced accuracy on validation set [6] was 0.8592 and Mean AUC was 0.9932. This methodology enabled us to improve model accuracy across a wide range of criteria, greatly surpassing all other methods.Additionally, our team capsule commandos achieved 7th place ranking with a test set[7] performance of Mean AUC: 0.7314 and balanced accuracy: 0.3235

eess.IV↗

arXiv2024

AutoSplat: Constrained Gaussian Splatting for Autonomous Driving Scene Reconstruction

Realistic scene reconstruction and view synthesis are essential for advancing autonomous driving systems by simulating safety-critical scenarios. 3D Gaussian Splatting excels in real-time rendering and static scene reconstructions but struggles with modeling driving scenarios due to complex backgrounds, dynamic objects, and sparse views. We propose AutoSplat, a framework employing Gaussian splatting to achieve highly realistic reconstructions of autonomous driving scenes. By imposing geometric constraints on Gaussians representing the road and sky regions, our method enables multi-view consistent simulation of challenging scenarios including lane changes. Leveraging 3D templates, we introduce a reflected Gaussian consistency constraint to supervise both the visible and unseen side of foreground objects. Moreover, to model the dynamic appearance of foreground objects, we estimate residual spherical harmonics for each foreground Gaussian. Extensive experiments on Pandaset and KITTI demonstrate that AutoSplat outperforms state-of-the-art methods in scene reconstruction and novel view synthesis across diverse driving scenarios. Visit our project page at https://autosplat.github.io/.

cs.CV↗

arXiv2024

Analysis of a Modular Autonomous Driving Architecture: The Top Submission to CARLA Leaderboard 2.0 Challenge

In this paper we present the architecture of the Kyber-E2E submission to the map track of CARLA Leaderboard 2.0 Autonomous Driving (AD) challenge 2023, which achieved first place. We employed a modular architecture for our solution consists of five main components: sensing, localization, perception, tracking/prediction, and planning/control. Our solution leverages state-of-the-art language-assisted perception models to help our planner perform more reliably in highly challenging traffic scenarios. We use open-source driving datasets in conjunction with Inverse Reinforcement Learning (IRL) to enhance the performance of our motion planner. We provide insight into our design choices and trade-offs made to achieve this solution. We also explore the impact of each component in the overall performance of our solution, with the intent of providing a guideline where allocation of resources can have the greatest impact.

cs.AI↗

arXiv2021

Augmenting Imitation Experience via Equivariant Representations

The robustness of visual navigation policies trained through imitation often hinges on the augmentation of the training image-action pairs. Traditionally, this has been done by collecting data from multiple cameras, by using standard data augmentations from computer vision, such as adding random noise to each image, or by synthesizing training images. In this paper we show that there is another practical alternative for data augmentation for visual navigation based on extrapolating viewpoint embeddings and actions nearby the ones observed in the training data. Our method makes use of the geometry of the visual navigation problem in 2D and 3D and relies on policies that are functions of equivariant embeddings, as opposed to images. Given an image-action pair from a training navigation dataset, our neural network model predicts the latent representations of images at nearby viewpoints, using the equivariance property, and augments the dataset. We then train a policy on the augmented dataset. Our simulation results indicate that policies trained in this way exhibit reduced cross-track error, and require fewer interventions compared to policies trained using standard augmentation methods. We also show similar results in autonomous visual navigation by a real ground robot along a path of over 500m.

cs.CV↗

arXiv2021

Good speciation and endogenous business cycles in a constraint satisfaction macroeconomic model

We introduce a prototype agent-based model of the macroeconomy, with budgetary constraints at its core. The model is related to a class of constraint satisfaction problems (CSPs), which has been thoroughly investigated in computer science. The CSP paradigm allows us to propose an alternative price-setting mechanism: given agents' preferences and budgets, what set of prices satisfies the maximum number of agents? Such an approach permits the coupling of production and output within the economy to the allowed level of debt in a simplified framework. Within our model, we identify three different regimes upon varying the amount of debt that each agent can accumulate before defaulting. In presence of a very loose constraint on debt, endogenous crises leading to waves of synchronized bankruptcies are present. In the opposite regime of very tight debt constraining, the bankruptcy rate is extremely high and the economy remains structure-less. In an intermediate regime, the economy is stable with very low bankruptcy rate and no aggregate-level crises. This third regime displays a rich phenomenology:the system spontaneously and dynamically self-organizes in a set of cheap and expensive goods (i.e. some kind of "speciation"), with switches triggered by random fluctuations and feedback loops. Our analysis confirms the central role that debt levels play in the stability of the economy. More generally, our model shows that constraints at the individual scale can generate highly complex patterns at the aggregate level.

econ.GN↗

arXiv2021

V-, U-, L-, or W-shaped economic recovery after COVID: Insights from an Agent Based Model

We discuss the impact of a Covid-19--like shock on a simple model economy, described by the previously developed Mark-0 Agent-Based Model. We consider a mixed supply and demand shock, and show that depending on the shock parameters (amplitude and duration), our model economy can display V-shaped, U-shaped or W-shaped recoveries, and even an L-shaped output curve with permanent output loss. This is due to the economy getting trapped in a self-sustained "bad" state. We then discuss two policies that attempt to moderate the impact of the shock: giving easy credit to firms, and the so-called helicopter money, i.e. injecting new money into the households savings. We find that both policies are effective if strong enough. We highlight the potential danger of terminating these policies too early, although inflation is substantially increased by lax access to credit. Finally, we consider the impact of a second lockdown. While we only discuss a limited number of scenarios, our model is flexible and versatile enough to accommodate a wide variety of situations, thus serving as a useful exploratory tool for a qualitative, scenario-based understanding of post-Covid recovery. The corresponding code is available on-line.

econ.GN↗

arXiv2020

Low reflection at zero or low-energies in the well-barrier scattering potentials

Probability of reflection $R(E)$ off a finite attractive scattering potential at zero or low energies is ordinarily supposed to be 1. However, a fully attractive potential presents a paradoxical result that $R(0)=0$ or $R(0)<1$, when an effective parameter $q$ of the potential admits special discrete values. Here, we report another class of finite potentials which are well-barrier (attractive-repulsive) type and which can be made to possess much less reflection at zero and low energies for a band of low values of $q$. These well-barrier potentials have only two real turning points for $E \in(V_{min}, V_{max})$, excepting $E=0$. We present two exactly solvable and two numerically solved models to confirm this phenomenon.

quant-ph↗

arXiv2019

Self-planting: digging holes in rough landscapes

Motivated by a potential application in economics, we investigate a simple dynamical scheme to produce planted solutions in optimization problems with continuous variables. We consider the perceptron model as a prototypical model. Starting from random input patterns and perceptron weights, we find a locally optimal assignment of weights by gradient descent; we then remove misclassified patterns (if any), and replace them by new, randomly extracted patterns. This "remove and replace" procedure is iterated until perfect classification is achieved. We call this procedure "self-planting" because the "planted" state is not pre-assigned but results from a co-evolution of weights and patterns. We find an algorithmic phase transition separating a region in which self-planting is efficiently achieved from a region in which it takes exponential time in the system size. We conjecture that this transition might exist in a broad class of similar problems.

cond-mat.dis-nn↗

arXiv2018

Facial Recognition with Encoded Local Projections

Encoded Local Projections (ELP) is a recently introduced dense sampling image descriptor which uses projections in small neighbourhoods to construct a histogram/descriptor for the entire image. ELP has shown to be as accurate as other state-of-the-art features in searching medical images while being time and resource efficient. This paper attempts for the first time to utilize ELP descriptor as primary features for facial recognition and compare the results with LBP histogram on the Labeled Faces in the Wild dataset. We have evaluated descriptors by comparing the chi-squared distance of each image descriptor versus all others as well as training Support Vector Machines (SVM) with each feature vector. In both cases, the results of ELP were better than LBP in the same sub-image configuration.

cs.CV↗

arXiv2017

Dirichlet spectrum of the paradigm model of complex PT-symmetric potential: $V(x)=-(ix)^N$

So far the spectra $E_n(N)$ of the paradigm model of complex PT(Parity-Time)-symmetric potential $V_{BB}(x,N)=-(ix)^N$ is known to be analytically continued for $N > 4$. Consequently, the well known eigenvalues of the Hermitian cases ($N=6,10$) cannot be recovered. Here, we illustrate Kato's theorem that even if a Hamiltonian $H(λ)$ is an analytic function of a real parameter $λ$, its eigenvalues $E_n(λ)$ may not be analytic at finite number of Isolated Points (IPs). In this light, we present the Dirichlet spectra $E_n(N)$ of $V_{BB}(x,N)$ for $2\le N<12$ using the numerical integration of Schr{ö}dinger equation with $ψ(x=\pm \infty)=0$ and the diagonalization of $H=p^2/2μ+V_{BB}(x,N)$ in the harmonic oscillator basis. We show that these real discrete spectra are consistent with the most simple two-turning point CWKB (C refers to complex turning points) method provided we choose the maximal turning points (MxTP) [$-a+ib,a+ib, a, b \in {\cal R}$] such that $|a|$ is the largest for a given energy among all (multiple) turning points. We find that $E_n(N)$ are continuous function of $N$ but non-analytic (their first derivative is discontinuous) at IPs $N=4,8$; where the Dirichlet spectrum is null (as $V_{BB}$ becomes a Hermitian flat-top potential barrier). At $N=6$ and $10$, $V_{BB}(x,N)$ becomes a Hermitian well and we recover its well known eigenvalues.

quant-ph↗

arXiv2017

Supersymmetric partner potentials arising from nodeless half bound states

A Half Bound State (HBS) $ψ_*(x)$ can be defined as a single, conditional, zero-energy, continuous solution of the one dimensional Schr{ö}dinger equation for a scattering potential well $V(x)$ ($s.t ~ V(\pm \infty)=0$). The non-normalizable and solitary HBS of a potential satisfies Neumann boundary condition that $ψ'_*(\pm \infty)=0$ and it can have $n$ (= 0,1,2,...) number of nodes indicating $n$ number of bound states in $V(x)$ below $E=0$. Here we show that starting with a nodeless HBS, we can construct a (supersymmetric) pair of finite potentials (well, double wells, well-barrier): $V_{\pm}(x)$ having no bound state and they enclose positive area on $x$-axis. On the contrary their negative counterparts $(-cV_{\pm}(x)),c>0$ do have at least one bound state for any arbitrary positive value of $c$. Furthermore, $c V_{\pm}(x),~ c >0$ which binds positive area on x-axis in conformity with Simon's theorem can have at least one bound state only conditionally for instance when $c>1$ or $c>>1$.

quant-ph↗

arXiv2015

Real discrete spectrum of complex PT-symmetric scattering potentials

We investigate the parametric evolution of the real discrete spectrum of several complex PT symmetric scattering potentials of the type $V(x)=-V_1 F_e(x) + i V_2 F_o(x), V_1>0, F_e(x)>0$ by varying $V_2$ slowly. Here $e,o$ stand for even and odd parity and $F_{e,o}(\pm \infty)=0$. Unlike the case of Scarf II potential, we find a general absence of the recently explored accidental (real to real) crossings of eigenvalues in these scattering potentials. On the other hand, we find a general presence of coalescing of real pairs of eigenvalues to the complex conjugate pairs at a finite number of exceptional points. We attribute such coalescings of eigenvalues to the presence of a finite barrier (on the either side of $x=0$ ) which has been linked to a recent study of stokes phenomenon in the complex PT-symmetric potentials.

quant-ph↗

arXiv2015

Avoided crossings of energy levels in one-dimension

Due to the absence of degeneracy in one dimension, when a parameter, $λ$, of a potential is varied slowly the discrete energy eigenvalue curves, $E_n(λ)$, cannot cross but they are allowed to come quite close and diverge from each other. This phenomena is called avoided crossing of energy levels. The parametric evolution of eigenvalues of the generally known one dimensional potentials do not display avoided crossings and on the other hand some complicated and analytically unsolvable models do exhibit this. Here, we show that this interesting spectral property can be found in simple one-dimensional double-wells when width of one of them is varied slowly.

quant-ph↗