arXiv subjects

Neha Gupta

Publications and source records attributed to Neha Gupta.

At least 19 recordsLinked to original sources

arXiv2026

LLMs Get Smarter from Targeted Synthetic Multilingual Data

Language-specific competency (LSC) is the phenomenon of a language model performing better or worse depending on the language of the prompt. In other words, a language model outputs different (and potentially incorrect) responses to the same semantic query when prompted in different languages. Prior work attributes this to an internal misalignment of semantic representation across languages. Currently, there are two main approaches to address LSC in the literature: (1) routing all queries through English, improving performance, but limiting language expressivity to English; or (2) training on language-balanced data, equalizing model performance across languages, but reducing overall performance. In this work, we take a data centric perspective and introduce HOTFIXR: Hardness Optimized Training data For Improving X-Lingual Reasoning. It is a data generation framework that uses models to probe and learn a student model's multilingual weaknesses, and generates data to mitigate them. HOTFIXR can generate multilingual synthetic training data that can improve multilingual performance. We evaluate on three in-distribution tasks, three out-of-distribution tasks, and four out-of-distribution languages. On average, HOTFIXR (1) improves in-distribution performance by 6.2%, (2) reduces catastrophic forgetting (induced by fine-tuning) on OOD tasks by 3.7%, and (3) on OOD languages by 7.1%. Overall, as many real-world applications requires multilingual LLMs, our work contributes to the efforts of making LLMs multilingually proficient. We will release code upon acceptance.

cs.CL↗

arXiv2026

Spillover-Informed Network Architecture for Global Volatility Forecasting

Spillover of volatility shocks across borders during turbulent periods makes accurate equity market volatility forecasts especially critical for risk management, derivatives pricing, and regulatory capital. In this paper, we examine whether volatility forecasts improve when models incorporate information on how markets are connected, and whether the choice of connection measure matters. Using daily data on 29 equity indices from every major region over 2015-2025, we let each market's forecast draw on the recent volatility of the markets linked to it, with the strength of each link set either by geography, return correlation, or the Diebold-Yilmaz (DY) spillover network estimated from the data. The spillover-informed forecasting model achieved an approximately 13\% reduction in out-of-sample QLIKE loss relative to the standard Heterogeneous AutoRegressive benchmark ($p<0.001$) and attained the highest model confidence set $p$-value among the models considered. Several benchmark models that do not explicitly incorporate cross-market structure were excluded from the 90\% model confidence set. The greatest improvements were observed during periods of elevated market stress: the COVID-19 crash, the Russian invasion of Ukraine, and the $2025$ US tariff shock, spillover-informed forecasts achieved approximately 21\% lower QLIKE loss than an otherwise identical network-blind model ($p = 0.018$), with no significant performance loss during calm periods. The advantage is economically material: a volatility-targeting investor would pay $322$-$373$ basis points per year, net of transaction costs, for spillover-informed forecasts, versus an insignificant $85$-$102$ basis point for the alternatives. Finally, when the model is left to infer connections on its own, it recovers the DY network from the forecast objective alone (permutation $p=0.0005$).

stat.AP↗

arXiv2026

NeuroMem-FHP: A Likelihood-Free Deep Learning Framework for Parameter Estimation of Fractional Hawkes Process

In this paper, we propose deep learning based NeuroMem-FHP framework for estimating the parameters of the fractional Hawkes process (FHP), a self-exciting point process that captures long-range dependence through a fractional Mittag-Leffler excitation kernel. Two neural architectures, namely a Long Short-Term Memory (LSTM) network and a Transformer, are developed to estimate the model parameters $(\mu,\gamma,\alpha,\beta)$ directly from sequences of inter-arrival times without requiring computationally intensive likelihood optimization. Experiments on synthetic data that both neural models significantly outperform the classical Maximum Likelihood Estimation (MLE) method, with the Transformer achieving the highest estimation accuracy (MSE = $0.1634$), followed by the LSTM (MSE = $0.1752$), compared to MLE (MSE = $2.8032$). An ablation study further examines the effects of key hyperparameters on model performance. The proposed framework is also on two real-world high-frequency datasets, namely AAPL NBBO transaction data and Montgomery County 911 emergency call records. Using a predictive validation approach, event sequences simulated from the estimated parameters closely reproduce the empirical distribution, tail behavior, and temporal dependence structure of the observed data. These results demonstrate that Transformer-based parameter estimation provides an accurate and efficient alternative to conventional estimation techniques for FHP and offers a promising framework for modeling event-driven systems with long-memory dynamics.

cs.LG↗

arXiv2026

Solving the Post-Quantum Control Plane Bottleneck: Energy-Aware Cryptographic Scheduling in Open RAN

The Open Radio Access Network (O-RAN) offers flexibility and innovation but introduces unique security vulnerabilities, particularly from cryptographically relevant quantum computers. While Post-Quantum Cryptography (PQC) is the primary scalable defence, its computationally intensive handshakes create a significant bottleneck for the RAN control plane, posing sustainability challenges. This paper proposes an energy-aware framework to solve this PQC bottleneck, ensuring quantum resilience without sacrificing operational energy efficiency. The system employs an O-RAN aligned split: a Crypto Policy rApp residing in the Non-Real-Time (Non-RT) RIC defines the strategic security envelope (including PQC suites), while a Security Operations Scheduling (SOS) xApp in the Near-RT RIC converts these into tactical timing and placement intents. Cryptographic enforcement remains at standards-compliant endpoints: the Open Fronthaul utilizes Media Access Control Security (MACsec) at the O-DU/O-RU, while the xhaul (midhaul and backhaul) utilizes IP Security (IPsec) at tunnel terminators. The SOS xApp reduces PQC overhead by batching non-urgent handshakes, prioritizing session resumption, and selecting parameters that meet slice SLAs while minimizing joules per secure connection. We evaluate the architecture via a Discrete-Event Simulation (DES) using 3GPP-aligned traffic profiles and verified hardware benchmarks from literature. Results show that intelligent scheduling can reduce per-handshake energy by approximately 60 percent without violating slice latency targets.

cs.CR↗

arXiv2026

Ministral 3

We introduce the Ministral 3 series, a family of parameter-efficient dense language models designed for compute and memory constrained applications, available in three model sizes: 3B, 8B, and 14B parameters. For each model size, we release three variants: a pretrained base model for general-purpose use, an instruction finetuned, and a reasoning model for complex problem-solving. In addition, we present our recipe to derive the Ministral 3 models through Cascade Distillation, an iterative pruning and continued training with distillation technique. Each model comes with image understanding capabilities, all under the Apache 2.0 license.

cs.CL↗

arXiv2025

When Actions Teach You to Think: Reasoning-Action Synergy via Reinforcement Learning in Conversational Agents

Supervised fine-tuning (SFT) has emerged as one of the most effective ways to improve the performance of large language models (LLMs) in downstream tasks. However, SFT can have difficulty generalizing when the underlying data distribution changes, even when the new data does not fall completely outside the training domain. Recent reasoning-focused models such as o1 and R1 have demonstrated consistent gains over their non-reasoning counterparts, highlighting the importance of reasoning for improved generalization and reliability. However, collecting high-quality reasoning traces for SFT remains challenging -- annotations are costly, subjective, and difficult to scale. To address this limitation, we leverage Reinforcement Learning (RL) to enable models to learn reasoning strategies directly from task outcomes. We propose a pipeline in which LLMs generate reasoning steps that guide both the invocation of tools (e.g., function calls) and the final answer generation for conversational agents. Our method employs Group Relative Policy Optimization (GRPO) with rewards designed around tool accuracy and answer correctness, allowing the model to iteratively refine its reasoning and actions. Experimental results demonstrate that our approach improves both the quality of reasoning and the precision of tool invocations, achieving a 1.5% relative improvement over the SFT model (trained without explicit thinking) and a 40% gain compared to the base of the vanilla Qwen3-1.7B model. These findings demonstrate the promise of unifying reasoning and action learning through RL to build more capable and generalizable conversational agents.

cs.CL↗

arXiv2025

NeuroMemFPP: A recurrent neural approach for memory-aware parameter estimation in fractional Poisson process

In this paper, we propose a recurrent neural network (RNN)-based framework for estimating the parameters of the fractional Poisson process (FPP), which models event arrivals with memory and long-range dependence. The Long Short-Term Memory (LSTM) network estimates the key parameters $\mu >0$ and $\beta \in(0,1)$ from sequences of inter-arrival times, effectively modeling their temporal dependencies. Our experiments on synthetic data show that the proposed approach reduces the mean squared error (MSE) by about 55.3\% compared to the traditional method of moments (MOM) and performs reliably across different training conditions. We tested the method on two real-world high-frequency datasets: emergency call records from Montgomery County, PA, and AAPL stock trading data. The results show that the LSTM can effectively track daily patterns and parameter changes, indicating its effectiveness on real-world data with complex time dependencies.

cs.LG↗

arXiv2025

Aurora: Architecting Argonne's First Exascale Supercomputer for Accelerated Scientific Discovery

Aurora is Argonne National Laboratory's pioneering Exascale supercomputer, designed to accelerate scientific discovery with cutting-edge architectural innovations. Key new technologies include the Intel(TM) Xeon(TM) Data Center GPU Max Series (code-named Sapphire Rapids) with support for High Bandwidth Memory (HBM), alongside the Intel(TM) Data Center GPU Max Series (code-named Ponte Vecchio) on each compute node. Aurora also integrates the Distributed Asynchronous Object Storage (DAOS), a novel exascale storage solution, and leverages Intel's oneAPI programming environment. This paper presents an in-depth exploration of Aurora's node architecture, the HPE Slingshot interconnect, the supporting software ecosystem, and DAOS. We provide insights into standard benchmark performance and applications readiness efforts via Aurora's Early Science Program and the Exascale Computing Project.

cs.DC↗

arXiv2025

DySK-Attn: A Framework for Efficient, Real-Time Knowledge Updating in Large Language Models via Dynamic Sparse Knowledge Attention

Large Language Models (LLMs) suffer from a critical limitation: their knowledge is static and quickly becomes outdated. Retraining these massive models is computationally prohibitive, while existing knowledge editing techniques can be slow and may introduce unforeseen side effects. To address this, we propose DySK-Attn, a novel framework that enables LLMs to efficiently integrate real-time knowledge from a dynamic external source. Our approach synergizes an LLM with a dynamic Knowledge Graph (KG) that can be updated instantaneously. The core of our framework is a sparse knowledge attention mechanism, which allows the LLM to perform a coarse-to-fine grained search, efficiently identifying and focusing on a small, highly relevant subset of facts from the vast KG. This mechanism avoids the high computational cost of dense attention over the entire knowledge base and mitigates noise from irrelevant information. We demonstrate through extensive experiments on time-sensitive question-answering tasks that DySK-Attn significantly outperforms strong baselines, including standard Retrieval-Augmented Generation (RAG) and model editing techniques, in both factual accuracy for updated knowledge and computational efficiency. Our framework offers a scalable and effective solution for building LLMs that can stay current with the ever-changing world.

cs.CL↗

arXiv2025

Devstral: Fine-tuning Language Models for Coding Agent Applications

We introduce Devstral-Small, a lightweight open source model for code agents with the best performance among models below 100B size. In this technical report, we give an overview of how we design and develop a model and craft specializations in agentic software development. The resulting model, Devstral-Small is a small 24B model, fast and easy to serve. Despite its size, Devstral-Small still attains competitive performance compared to models more than an order of magnitude larger.

cs.SE↗

arXiv2025

Voxtral

We present Voxtral Mini and Voxtral Small, two multimodal audio chat models. Voxtral is trained to comprehend both spoken audio and text documents, achieving state-of-the-art performance across a diverse range of audio benchmarks, while preserving strong text capabilities. Voxtral Small outperforms a number of closed-source models, while being small enough to run locally. A 32K context window enables the model to handle audio files up to 40 minutes in duration and long multi-turn conversations. We also contribute three benchmarks for evaluating speech understanding models on knowledge and trivia. Both Voxtral models are released under Apache 2.0 license.

cs.SD↗

arXiv2025

A Unified Approach to Calculating Sylvester Sums

In the context of the Frobenius coin problem, given two relatively prime positive integers $a$ and $b$, the set of nonrepresentable numbers consists of positive integers that cannot be expressed as nonnegative integer combination of $a$ and $b$. This work provides a formula for calculating the power sums of all nonrepresentable numbers, also known as the Sylvester sums. Although alternative formulas exist in the literature, our approach is based on an elementary observation. We consider the set of natural numbers from $1$ to $ab - 1$ and compute their total sum in two distinct ways, which leads naturally to the desired Sylvester sums. This method connects an analytic identity with a combinatorial viewpoint, giving a new way to understand these classical quantities. Furthermore, in this paper, we establish a criterion using the division algorithm to determine whether a given positive integer is nonrepresentable.

math.NT↗

arXiv2025

Magistral

We introduce Magistral, Mistral's first reasoning model and our own scalable reinforcement learning (RL) pipeline. Instead of relying on existing implementations and RL traces distilled from prior models, we follow a ground up approach, relying solely on our own models and infrastructure. Notably, we demonstrate a stack that enabled us to explore the limits of pure RL training of LLMs, present a simple method to force the reasoning language of the model, and show that RL on text data alone maintains most of the initial checkpoint's capabilities. We find that RL on text maintains or improves multimodal understanding, instruction following and function calling. We present Magistral Medium, trained for reasoning on top of Mistral Medium 3 with RL alone, and we open-source Magistral Small (Apache 2.0) which further includes cold-start data from Magistral Medium.

cs.CL↗

arXiv2025

Pre-Act: Multi-Step Planning and Reasoning Improves Acting in LLM Agents

The ReAct (Reasoning + Action) capability in large language models (LLMs) has become the foundation of modern agentic systems. Recent LLMs, such as DeepSeek-R1 and OpenAI o1/o3, exemplify this by emphasizing reasoning through the generation of ample intermediate tokens, which help build a strong premise before producing the final output tokens. In this paper, we introduce Pre-Act, a novel approach that enhances the agent's performance by creating a multi-step execution plan along with the detailed reasoning for the given user input. This plan incrementally incorporates previous steps and tool outputs, refining itself after each step execution until the final response is obtained. Our approach is applicable to both conversational and non-conversational agents. To measure the performance of task-oriented agents comprehensively, we propose a two-level evaluation framework: (1) turn level and (2) end-to-end. Our turn-level evaluation, averaged across five models, shows that our approach, Pre-Act, outperforms ReAct by 70% in Action Recall on the Almita dataset. While this approach is effective for larger models, smaller models crucial for practical applications, where latency and cost are key constraints, often struggle with complex reasoning tasks required for agentic systems. To address this limitation, we fine-tune relatively small models such as Llama 3.1 (8B & 70B) using the proposed Pre-Act approach. Our experiments show that the fine-tuned 70B model outperforms GPT-4, achieving a 69.5% improvement in action accuracy (turn-level) and a 28% improvement in goal completion rate (end-to-end) on the Almita (out-of-domain) dataset.

cs.AI↗

arXiv2025

Noncentral moderate deviations for time-changed multivariate L\'evy processes with linear combinations of inverse stable subordinators

The term noncentral moderate deviations is used in the literature to mean a class of large deviation principles that, in some sense, fills the gap between the convergence in probability to a constant (governed by a reference large deviation principle) and a weak convergence to a non-Gaussian (and non-degenerating) distribution. Some noncentral moderate deviation results in the literature concern time-changed univariate L\'evy processes, where the time-changes are given by inverse stable subordinators. In this paper we present analogue results for multivariate L\'evy processes; in particular the random time-changes are suitable linear combinations of independent inverse stable subordinators.

math.PR↗

arXiv2025

Geometrical subordinated Poisson processes and its extensions

In this paper, we study a generalized version of the Poisson-type process by time-changing it with the geometric counting process. Our work generalizes the work done by Meoli (2023) \cite{meoli2023some}. We defined the geometric subordinated Poisson process (GSPP), the geometric subordinated compound Poisson process (GSCPP) and the geometric subordinated multiplicative Poisson process (GSMPP) by time-changing the subordinated Poisson process, subordinated compound Poisson process and subordinated multiplicative Poisson process with the geometric counting process, respectively. We derived several distributional properties and many special cases from the above-mentioned processes. We calculate the asymptotic behavior of the correlation structure. We have discussed applications of time-changed generalized compound Poisson in shock modelling.

math.PR↗

arXiv2024

The Frobenius number for the triple of the 2-step star numbers

In this paper, we give closed form expressions of the Frobenius number for the triple of the $2$-step star numbers $an(n-2) + 1$ for an integer $a \geq 4$. These numbers have been studied from different aspects for some $a$'s. These numbers can also be considered as variations of the well known star numbers of the form $6n(n-1) + 1$. We also give closed form expressions of the Sylvester number (genus) for the triple of the $2$-step star numbers.

math.CO↗

arXiv2024

Hawkes process with tempered Mittag-Leffler kernel

In this paper, we propose an extension of the Hawkes process by incorporating a kernel based on the tempered Mittag-Leffler distribution. This is the generalization of the work presented in [10]. We derive analytical results for the expectation of the conditional intensity and the expected number of events in the counting process. Additionally, we investigate the limiting behavior of the expectation of the conditional intensity. Finally, we present an empirical comparison of the studied process with its limiting special cases.

math.PR↗