arXiv subjects

Sukruta Midigeshi

Publications and source records attributed to Sukruta Midigeshi.

2 recordsLinked to original sources

arXiv2026

Packing Linear Programs and Fractional Knapsack using Comparison Oracles

We study the problem of recovering the objective of a packing linear program when the algorithm accesses only comparison information about optimal solutions under varying constraint matrices. Motivated by optimization with comparison oracles (Cohen-Addad et al., STOC 2026) and preference feedback (Kaufmann et al., TMLR 2025), this strengthens inverse-optimization frameworks by replacing direct observations of optimal solutions with ordinal queries. We focus on the fractional knapsack problem, where the packing linear program (LP) has a single budget constraint specified by item prices, and the objective is determined by item values. This captures monopoly-pricing where a seller infers a buyer's unknown valuations for divisible items from comparison information. The algorithm queries an oracle with two price vectors, returning which optimal solution has the larger total packing or objective value. Such oracles abstract discrete-choice surveys of buyers choosing between differently priced alternatives. For fractional knapsack, we develop a polynomial-time algorithm recovering item values up to scale using $O(n \log(1/\delta)+B^2)$ comparison queries, where $n$ is the number of items, $B$ is the knapsack capacity, and $\delta$ is the value grid resolution. We complement this with an $\Omega(n \log(1/\delta))$ lower bound. A key insight is that in the comparison-oracle model, fractional knapsack is as general as packing LPs. Our algorithm solves the packing setting by treating a constraint matrix row as the price vector and zeroing the rest. The $\Omega(n \log(1/\delta))$ lower bound continues to hold for packing LPs, making our upper bound essentially best possible, up to a linear-factor gap. Finally, we extend our algorithm to profit-maximization, yielding a comparison-oracle analogue of the revealed-preference result of Amin et al. (AAAI 2015).

cs.DS↗

arXiv2025

Characterizing Deep Research: A Benchmark and Formal Definition

Information tasks such as writing surveys or analytical reports require complex search and reasoning, and have recently been grouped under the umbrella of \textit{deep research} -- a term also adopted by recent models targeting these capabilities. Despite growing interest, the scope of the deep research task remains underdefined and its distinction from other reasoning-intensive problems is poorly understood. In this paper, we propose a formal characterization of the deep research (DR) task and introduce a benchmark to evaluate the performance of DR systems. We argue that the core defining feature of deep research is not the production of lengthy report-style outputs, but rather the high fan-out over concepts required during the search process, i.e., broad and reasoning-intensive exploration. To enable objective evaluation, we define DR using an intermediate output representation that encodes key claims uncovered during search-separating the reasoning challenge from surface-level report generation. Based on this formulation, we propose a diverse, challenging benchmark LiveDRBench with 100 challenging tasks over scientific topics (e.g., datasets, materials discovery, prior art search) and public interest events (e.g., flight incidents, movie awards). Across state-of-the-art DR systems, F1 score ranges between 0.02 and 0.72 for any sub-category. OpenAI's model performs the best with an overall F1 score of 0.55. Analysis of reasoning traces reveals the distribution over the number of referenced sources, branching, and backtracking events executed by current DR systems, motivating future directions for improving their search mechanisms and grounding capabilities. The benchmark is available at https://github.com/microsoft/LiveDRBench.

cs.CL↗