arXiv subjects

Swati Gupta

Publications and source records attributed to Swati Gupta.

At least 19 recordsLinked to original sources

arXiv2026

Why Global LLM Leaderboards Are Misleading: Small Portfolios for Heterogeneous Supervised ML

Ranking LLMs via pairwise human feedback underpins current leaderboards for open-ended tasks, such as creative writing and problem-solving. We analyze ~89K comparisons in 116 languages from 52 LLMs from Arena, and show that the best-fit global Bradley-Terry (BT) ranking is misleading. Nearly 2/3 of the decisive votes cancel out, and even the top 50 models according to the global BT ranking are statistically indistinguishable (pairwise win probabilities are at most 0.53 within the top 50 models). We trace this failure to strong, structured heterogeneity of opinions across language, task, and time. Moreover, we find an important characteristic - *language* plays a key role. Grouping by language (and families) increases the agreement of votes massively, resulting in two orders of magnitude higher spread in the ELO scores (i.e., very consistent rankings). What appears as global noise is in fact a mixture of coherent but conflicting subpopulations. To address such heterogeneity in supervised machine learning, we introduce the framework of $(\lambda, \nu)$-portfolios, which are small sets of models that achieve a prediction error at most $\lambda$, "covering" at least a $\nu$ fraction of users. We formulate this as a variant of the set cover problem and provide guarantees using the VC dimension of the underlying set system. On the Arena data, our algorithms recover just 5 distinct BT rankings that cover over 96% of votes at a modest $\lambda$, compared to the 21% coverage by the global ranking. We also provide a portfolio of 6 LLMs that cover twice as many votes as the top-6 LLMs from a global ranking. We further construct portfolios for a classification problem on the COMPAS dataset using an ensemble of fairness-regularized classification models and show that these portfolios can be used to detect blind spots in the data, which might be of independent interest to policymakers.

cs.LG↗

arXiv2026

Many Preferences, Few Policies: Towards Scalable Language Model Personalization

The holy grail of LLM personalization is a single LLM for each user, perfectly aligned with that user's preferences. However, maintaining a separate LLM per user is impractical due to constraints on compute, memory, and system complexity. We address this challenge by developing a principled method for selecting a small portfolio of LLMs that captures representative behaviors across heterogeneous users. We model user preferences across multiple traits (e.g., safety, humor, brevity) through a multi-dimensional weight vector. Given reward functions across these dimensions, our algorithm PALM (Portfolio of Aligned LLMs) generates a small portfolio of LLMs such that, for any weight vector, the portfolio contains a near-optimal LLM for the corresponding scalarized objective. To the best of our knowledge, this is the first result that provides theoretical guarantees on both the size and approximation quality of LLM portfolios for personalization. It characterizes the trade-off between system cost and personalization, as well as the diversity of LLMs required to cover the landscape of user preferences. We provide empirical results that validate these guarantees and demonstrate greater output diversity over common baselines.

cs.CL↗

arXiv2026

Faster Parametric Submodular Function Minimization by Exploiting Duality

Let $f:2^{E} \rightarrow \mathbb{Z}_+$ be a submodular function on a ground set $E = [n]$, and let $P(f)$ denote its extended polymatroid. Given a direction $d \in \mathbb{Z}^n$ with at least one positive entry, the line search problem is to find the largest scalar $\lambda$ such that $\lambda d \in P(f)$. The best known strongly polynomial-time algorithm for this problem is based on the discrete Newton's method and requires $\tilde{O}(n^2 \log n)\cdot$ SFM time, where SFM is the time for exact submodular function minimization under the value oracle model. In this work, we study the first weakly polynomial-time algorithms for this problem. We reduce the number of calls to the exact submodular minimization oracle by exploiting a dual formulation of the parametric line search problem and recent advances in cutting plane methods. We obtain a running time of \[ O\bigl(n^2 \log(nM\|d\|_1)\cdot \text{EO} + n^3 \log(nM\|d\|_1)\bigr) + O(1)\cdot \text{SFM}, \] where $M = \|f\|_\infty$ and EO is the cost of evaluating $f$ at a set. Note that when $\log \|d\|_1 = O(\log (nM))$, this matches the current best weakly polynomial running time for submodular function minimization [Lee, Sidford, Wong '15], and therefore, one cannot hope to improve this running time. Our approach proceeds by deriving a dual formulation that minimizes the Lov\'asz extension $F$ over a hyperplane intersecting the unit hypercube, and then solving this dual problem approximately via cutting-plane methods, after which we round to the exact intersection using the integrality of $f$ and $d$.

math.OC↗

arXiv2026

Improved Regret Guarantees for Online Mirror Descent using a Portfolio of Mirror Maps

OMD and its variants give a flexible framework for OCO where the performance depends crucially on the choice of the mirror map. While the geometries underlying OPGD and OEG, both special cases of OMD, are well understood, it remains a challenging open question on how to construct an optimal mirror map for any given constrained set and a general family of loss functions, e.g., sparse losses. Motivated by parameterizing a near-optimal set of mirror maps, we consider a simpler question: is it even possible to obtain polynomial gains in regret by using mirror maps for geometries that interpolate between $L_1$ and $L_2$, which may not be possible by restricting to only OEG ($L_1$) or OPGD ($L_2$). Our main result answers this question positively. We show that mirror maps based on block norms adapt better to the sparsity of loss functions, compared to previous $L_p$ (for $p \in [1, 2]$) interpolations. In particular, we construct a family of online convex optimization instances in $\mathbb{R}^d$, where block norm-based mirror maps achieve a provable polynomial (in $d$) improvement in regret over OEG and OPGD for sparse loss functions. We then turn to the setting in which the sparsity level of the loss functions is unknown. In this case, the choice of geometry itself becomes an online decision problem. We first show that naively switching between OEG and OPGD can incur linear regret, highlighting the intrinsic difficulty of geometry selection. To overcome this issue, we propose a meta-algorithm based on multiplicative weights that dynamically selects among a family of uniform block norms. We show that this approach effectively tunes OMD to the sparsity of the losses, yielding adaptive regret guarantees. Overall, our results demonstrate that online mirror-map selection can significantly enhance the ability of OMD to exploit sparsity in online convex optimization.

math.OC↗

arXiv2026

AnyView: Synthesizing Any Novel View in Dynamic Scenes

Modern generative video models excel at producing convincing, high-quality outputs, but struggle to maintain multi-view and spatiotemporal consistency in highly dynamic real-world environments. In this work, we introduce \textbf{AnyView}, a diffusion-based video generation framework for \emph{dynamic view synthesis} with minimal inductive biases or geometric assumptions. We leverage multiple data sources with various levels of supervision, including monocular (2D), multi-view static (3D) and multi-view dynamic (4D) datasets, to train a generalist spatiotemporal implicit representation capable of producing zero-shot novel videos from arbitrary camera locations and trajectories. We evaluate AnyView on standard benchmarks, showing competitive results with the current state of the art, and propose \textbf{AnyViewBench}, a challenging new benchmark tailored towards \emph{extreme} dynamic view synthesis in diverse real-world scenarios. In this more dramatic setting, we find that most baselines drastically degrade in performance, as they require significant overlap between viewpoints, while AnyView maintains the ability to produce realistic, plausible, and spatiotemporally consistent videos when prompted from \emph{any} viewpoint. Results, data, code, and models can be viewed at: https://tri-ml.github.io/AnyView/

cs.CV↗

arXiv2025

Provably Small Portfolios for Multiobjective Optimization with Application to Subsidized Facility Location

Many multiobjective real-world problems, such as facility location and bus routing, become more complex when optimizing the priorities of multiple stakeholders. These are often modeled using infinite classes of objectives, e.g., $L_p$ norms over group distances induced by feasible solutions in a fixed domain. Traditionally, the literature has considered explicitly balancing `equity' (or min-max) and `efficiency' (or min-sum) objectives to capture this trade-off. However, the structure of solutions obtained by such modeling choices can be very different. Taking a solution-centric approach, we introduce the concept of provably small set of solutions $P$, called a {\it portfolio}, such that for every objective function $h(\cdot)$ in the given class $\mathbf{C}$, there exists some solution in $P$ which is an $\alpha$-approximation for $h(\cdot)$. Constructing such portfolios can help decision-makers understand the impact of balancing across multiple objectives. Given a finite set of base objectives $h_1, \ldots, h_N$, we give provable algorithms for constructing portfolios for (1) the class of conic combinations $\mathbf{C} = \{\sum_{j \in [N]}\lambda_j h_j: \lambda \ge 0\}$ and for (2) any class $\mathbf{C}$ of functions that interpolates monotonically between the min-sum efficiency objective (i.e., $h_1 + \ldots + h_N$) and the min-max equity objective (i.e., $\max_{j \in [N]} h_j$). Examples of the latter are $L_p$ norms and top-$\ell$ norms. As an application, we study the Fair Subsidized Facility Location (FSFL) problem, motivated by the crisis of medical deserts caused due to pharmacy closures. FSFL allows subsidizing facilities in underserved areas using revenue from profitable locations. We develop a novel bicriteria approximation algorithm and show a significant reduction of medical deserts across states in the U.S.

cs.DS↗

arXiv2025

Comparison of Hyperplane Rounding for Max-Cut and Quantum Approximate Optimization Algorithm over Certain Regular Graph Families

There is a strong interest in finding challenging instances of NP-hard problems, from the perspective of showing quantum advantage. Due to the limits of near-term NISQ devices, it is moreover useful if these instances are small. In this work, we identify two graph families ($|V|<1000$) on which the Goemans-Williamson algorithm for approximating the Max-Cut achieves at most a 0.912-approximation. We further show that, in comparison, a recent quantum algorithm, Quantum Approximate Optimization Algorithm (depth $p=1$), is a 0.592-approximation on Karloff instances in the limit ($n \to \infty$), and is at best a $0.894$-approximation on a family of strongly-regular graphs. We further explore construction of challenging instances computationally by perturbing edge weights, which may be of independent interest, and include these in the CI-QuBe github repository.

quant-ph↗

arXiv2025

A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation

Robot manipulation has seen tremendous progress in recent years, with imitation learning policies enabling successful performance of dexterous and hard-to-model tasks. Concurrently, scaling data and model size has led to the development of capable language and vision foundation models, motivating large-scale efforts to create general-purpose robot foundation models. While these models have garnered significant enthusiasm and investment, meaningful evaluation of real-world performance remains a challenge, limiting both the pace of development and inhibiting a nuanced understanding of current capabilities. In this paper, we rigorously evaluate multitask robot manipulation policies, referred to as Large Behavior Models (LBMs), by extending the Diffusion Policy paradigm across a corpus of simulated and real-world robot data. We propose and validate an evaluation pipeline to rigorously analyze the capabilities of these models with statistical confidence. We compare against single-task baselines through blind, randomized trials in a controlled setting, using both simulation and real-world experiments. We find that multi-task pretraining makes the policies more successful and robust, and enables teaching complex new tasks more quickly, using a fraction of the data when compared to single-task baselines. Moreover, performance predictably increases as pretraining scale and diversity grows. Project page: https://toyotaresearchinstitute.github.io/lbm1/

cs.RO↗

arXiv2025

Making a Case for Research Collaboration Between Artificial Intelligence and Operations Research Experts

In 2021, INFORMS, ACM SIGAI, and the Computing Community Consortium (CCC) hosted three workshops to explore synergies between Artificial Intelligence (AI) and Operations Research (OR) to improve decision-making. The workshops aimed to create a unified research vision for AI/OR collaboration, focusing on overcoming cultural differences and maximizing societal impact. The first two workshops addressed technological innovations, applications, and trustworthy AI development, while the final workshop highlighted specific areas for AI/OR integration. Participants discussed "Challenge Problems" and strategies for combining AI and OR techniques. This report outlines five key recommendations to enhance AI/OR collaboration: 1) Funding Opportunities, 2) Joint Education, 3) Long-term Research Programs, 4) Aligning Conferences/Journals, and 5) Benchmark Creation.

cs.CY↗

arXiv2025

Navigating the Social Welfare Frontier: Portfolios for Multi-objective Reinforcement Learning

In many real-world applications of reinforcement learning (RL), deployed policies have varied impacts on different stakeholders, creating challenges in reaching consensus on how to effectively aggregate their preferences. Generalized $p$-means form a widely used class of social welfare functions for this purpose, with broad applications in fair resource allocation, AI alignment, and decision-making. This class includes well-known welfare functions such as Egalitarian, Nash, and Utilitarian welfare. However, selecting the appropriate social welfare function is challenging for decision-makers, as the structure and outcomes of optimal policies can be highly sensitive to the choice of $p$. To address this challenge, we study the concept of an $\alpha$-approximate portfolio in RL, a set of policies that are approximately optimal across the family of generalized $p$-means for all $p \in [-\infty, 1]$. We propose algorithms to compute such portfolios and provide theoretical guarantees on the trade-offs among approximation factor, portfolio size, and computational efficiency. Experimental results on synthetic and real-world datasets demonstrate the effectiveness of our approach in summarizing the policy space induced by varying $p$ values, empowering decision-makers to navigate this landscape more effectively.

cs.LG↗

arXiv2024

Strategies for running the QAOA at hundreds of qubits

We explore strategies aimed at reducing the amount of computation, both quantum and classical, required to run the Quantum Approximate Optimization Algorithm (QAOA). First, following Wurtz et al. [Phys.Rev A 104:052419], we consider the standard QAOA with instance-independent "tree" parameters chosen in advance. These tree parameters are chosen to optimize the MaxCut expectation for large girth graphs. We provide extensive numerical evidence supporting the performance guarantee for tree parameters conjectured in [Phys.Rev A 103:042612] and see that the approximation ratios obtained with tree parameters are typically well beyond the conjectured lower bounds, often comparable to performing a full optimization. This suggests that in practice, the QAOA can achieve near-optimal performance without the need for parameter optimization. Next, we modify the warm-start QAOA of Tate et al. [Quantum 7:1121]. The starting state for the QAOA is now an optimized product state associated with a solution of the Goemans-Williamson (GW) algorithm. Surprisingly, the tree parameters continue to perform well for the warm-start QAOA. We find that for random 3-regular graphs with hundreds of vertices, the expected cut obtained by the warm-start QAOA at depth $p \gtrsim 3$ is comparable to that of the standard GW algorithm. Our numerics on random instances do not provide general performance guarantees but do provide substantial evidence that there exists a regime of instance sizes in which the QAOA finds good solutions at low depth without the need for parameter optimization. For each instance studied, we classically compute the expected size of the QAOA distribution of cuts; producing the actual cuts requires running on a quantum computer.

quant-ph↗

arXiv2024

Promise of Graph Sparsification and Decomposition for Noise Reduction in QAOA: Analysis for Trapped-Ion Compilations

We develop new approximate compilation schemes that significantly reduce the expense of compiling the Quantum Approximate Optimization Algorithm (QAOA) for solving the Max-Cut problem. Our main focus is on compilation with trapped-ion simulators using Pauli-$X$ operations and all-to-all Ising Hamiltonian $H_\text{Ising}$ evolution generated by Molmer-Sorensen or optical dipole force interactions, though some of our results also apply to standard gate-based compilations. Our results are based on principles of graph sparsification and decomposition; the former reduces the number of edges in a graph while maintaining its cut structure, while the latter breaks a weighted graph into a small number of unweighted graphs. Though these techniques have been used as heuristics in various hybrid quantum algorithms, there have been no guarantees on their performance, to the best of our knowledge. This work provides the first provable guarantees using sparsification and decomposition to improve quantum noise resilience and reduce quantum circuit complexity. For quantum hardware that uses edge-by-edge QAOA compilations, sparsification leads to a direct reduction in circuit complexity. For trapped-ion quantum simulators implementing all-to-all $H_\text{Ising}$ pulses, we show that for a $(1-\epsilon)$ factor loss in the Max-Cut approximation ($\epsilon>0)$, our compilations improve the (worst-case) number of $H_\text{Ising}$ pulses from $O(n^2)$ to $O(n\log(n/\epsilon))$ and the (worst-case) number of Pauli-$X$ bit flips from $O(n^2)$ to $O\left(\frac{n\log(n/\epsilon)}{\epsilon^2}\right)$ for $n$-node graphs. We demonstrate significant reductions in noise are obtained in our new compilation approaches using theory and numerical calculations for trapped-ion hardware. We anticipate these approximate compilation techniques will be useful tools in a variety of future quantum computing experiments.

quant-ph↗

arXiv2024

Equitably allocating wildfire resilience investments for power grids: The curse of aggregation and vulnerability indices

Social vulnerability indices have increased traction for guiding infrastructure investment decisions to prioritize communities that need these investments most. One such plan is the Biden-Harris Justice40 initiative, which aims to guide equitable infrastructure investments by ensuring that disadvantaged communities defined by the Climate & Economic Justice Screening Tool (CEJST) receive 40% of the total benefit realized by the investment. However, there is limited research on the practicality of applying vulnerability indices like the CEJST to real-world decision-making for policy outcomes. In this paper, we study this gap by examining the effectiveness of vulnerability indices in a case study focused on power shutoff and undergrounding decisions in wildfire-prone regions. Using a mixed-integer program and a high-fidelity synthetic transmission network in Texas, we model resource allocation policies inspired by Justice40 and evaluate their impact on reducing power outages and mitigating wildfire risk for vulnerable groups. Our analysis reveals that the Justice40 framework may fail to protect certain communities facing high wildfire risk. In our case study, we show that indigenous groups are particularly impacted. We posit that this outcome is likely due to information losses from data aggregation and the use of generalized vulnerability indices. Through the use of explicit group-level protections, we are able to bound the best possible outcome for population groups that are proportionally most affected.

math.OC↗

arXiv2023

Challenges and Opportunities in Quantum Optimization

Recent advances in quantum computers are demonstrating the ability to solve problems at a scale beyond brute force classical simulation. As such, a widespread interest in quantum algorithms has developed in many areas, with optimization being one of the most pronounced domains. Across computer science and physics, there are a number of different approaches for major classes of optimization problems, such as combinatorial optimization, convex optimization, non-convex optimization, and stochastic extensions. This work draws on multiple approaches to study quantum optimization. Provably exact versus heuristic settings are first explained using computational complexity theory - highlighting where quantum advantage is possible in each context. Then, the core building blocks for quantum optimization algorithms are outlined to subsequently define prominent problem classes and identify key open questions that, if answered, will advance the field. The effects of scaling relevant problems on noisy quantum devices are also outlined in detail, alongside meaningful benchmarking problems. We underscore the importance of benchmarking by proposing clear metrics to conduct appropriate comparisons with classical optimization techniques. Lastly, we highlight two domains - finance and sustainability - as rich sources of optimization problems that could be used to benchmark, and eventually validate, the potential real-world impact of quantum optimization.

quant-ph↗

arXiv2023

Balancing Notions of Equity: Trade-offs Between Fair Portfolio Sizes and Achievable Guarantees

Motivated by fairness concerns, we study the `portfolio problem': given an optimization problem with set $D$ of feasible solutions, a class $\mathbf{C}$ of fairness objective functions on $D$, and an approximation factor $\alpha \ge 1$, a set $X \subseteq D$ of feasible solutions is an $\alpha$-approximate portfolio if for each objective $f \in \mathbf{C}$, there is an $\alpha$-approximation for $f$ in $X$. Choosing the classes of top-$k$ norms, ordered norms, and symmetric monotonic norms as our equity objectives, we study the trade-off between the size $|X|$ of the portfolio and its approximation factor $\alpha$ for various combinatorial problems. For the problem of scheduling identical jobs on unidentical machines, we characterize this trade-off for ordered norms and give an exponential improvement in size for symmetric monotonic norms over the general upper bound. We generalize this result as the OrderAndCount framework that obtains an exponential improvement in portfolio sizes for covering polyhedra with a constant number of constraints. Our framework is based on a novel primal-dual counting technique that may be of independent interest. We also introduce a general IterativeOrdering framework for simultaneous approximations or portfolios of size $1$ for symmetric monotonic norms, which generalizes and extends existing results for problems such as scheduling, $k$-clustering, set cover, and routing.

cs.DS↗

arXiv2023

Improving Clinical Decision Support through Interpretable Machine Learning and Error Handling in Electronic Health Records

The objective of this work is to develop an Electronic Medical Record (EMR) data processing tool that confers clinical context to Machine Learning (ML) algorithms for error handling, bias mitigation and interpretability. We present Trust-MAPS, an algorithm that translates clinical domain knowledge into high-dimensional, mixed-integer programming models that capture physiological and biological constraints on clinical measurements. EMR data is projected onto this constrained space, effectively bringing outliers to fall within a physiologically feasible range. We then compute the distance of each data point from the constrained space modeling healthy physiology to quantify deviation from the norm. These distances, termed "trust-scores," are integrated into the feature space for downstream ML applications. We demonstrate the utility of Trust-MAPS by training a binary classifier for early sepsis prediction on data from the 2019 PhysioNet Computing in Cardiology Challenge, using the XGBoost algorithm and applying SMOTE for overcoming class-imbalance. The Trust-MAPS framework shows desirable behavior in handling potential errors and boosting predictive performance. We achieve an AUROC of 0.91 (0.89, 0.92 : 95% CI) for predicting sepsis 6 hours before onset - a marked 15% improvement over a baseline model trained without Trust-MAPS. Trust-scores emerge as clinically meaningful features that not only boost predictive performance for clinical decision support tasks, but also lend interpretability to ML models. This work is the first to translate clinical domain knowledge into mathematical constraints, model cross-vital dependencies, and identify aberrations in high-dimensional medical data. Our method allows for error handling in EMR, and confers interpretability and superior predictive power to models trained for clinical decision support.

cs.LG↗

arXiv2023

TACOS: Topology-Aware Collective Algorithm Synthesizer for Distributed Machine Learning

The surge of artificial intelligence, particularly large language models, has driven the rapid development of large-scale machine learning clusters. Executing distributed models on these clusters is often constrained by communication overhead, making efficient utilization of available network resources crucial. As a result, the routing algorithm employed for collective communications (i.e., collective algorithms) plays a pivotal role in determining overall performance. Unfortunately, existing collective communication libraries for distributed machine learning are limited by a fixed set of basic collective algorithms. This limitation hinders communication optimization, especially in modern clusters with heterogeneous and asymmetric topologies. Furthermore, manually designing collective algorithms for all possible combinations of network topologies and collective patterns requires heavy engineering and validation efforts. To address these challenges, this paper presents TACOS, an autonomous synthesizer capable of automatically generating topology-aware collective algorithms tailored to specific collective patterns and network topologies. TACOS is highly flexible, synthesizing an All-Reduce algorithm for a heterogeneous 128-NPU system in just 1.08 seconds, while achieving up to a 4.27x performance improvement over state-of-the-art synthesizers. Additionally, TACOS demonstrates better scalability with polynomial synthesis times, in contrast to NP-hard approaches which only scale to systems with tens of NPUs. TACOS can synthesize for 40K NPUs in just 2.52 hours.

cs.DC↗

arXiv2023

Artificial Intelligence/Operations Research Workshop 2 Report Out

This workshop Report Out focuses on the foundational elements of trustworthy AI and OR technology, and how to ensure all AI and OR systems implement these elements in their system designs. Four sessions on various topics within Trustworthy AI were held, these being Fairness, Explainable AI/Causality, Robustness/Privacy, and Human Alignment and Human-Computer Interaction. Following discussions of each of these topics, workshop participants also brainstormed challenge problems which require the collaboration of AI and OR researchers and will result in the integration of basic techniques from both fields to eventually benefit societal needs.

cs.AI↗