arXiv subjects

Pallavi Jain

Publications and source records attributed to Pallavi Jain.

At least 19 recordsLinked to original sources

arXiv2026

Reaching Fairness by Reallocating Goods

Fair allocation of indivisible goods has largely been studied under the assumption that no prior allocation exists. Motivated by practical settings with pre-existing (and possibly unfair) allocations, we study how to achieve fairness through limited reallocations. Building on recent work on reformability/reallocations, we consider three fairness notions---envy-freeness (EF), envy-freeness up to one good (EF1), and envy-freeness up to any good (EFX)---and optimize the number of goods reallocated. We analyze both the classical and parameterized complexity of these problems, providing a comprehensive analysis across multiple fairness notions.

cs.GT↗

arXiv2026

Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

Sparse autoencoders (SAEs) are increasingly used to extract interpretable concepts from vision and vision language models, yet existing evaluation methods largely rely on proxy metrics or qualitative inspection rather than measuring semantic correspondence. We present a human-grounded evaluation framework that quantifies alignment between SAE latents and human-annotated concepts, without requiring user studies, and validate this matching through targeted attribute perturbations. To enable this intervention-style evaluation in vision, we construct synCUB and synCOCO, synthetic benchmarks of paired images that differ in exactly one attribute. We introduce Fully-Binary Matching Pursuit (FBMP), a coalition-based matching procedure that supports many-to-one mappings between SAE latents and annotated concepts, and consistently outperforms one-to-one baselines. For functional validation, we propose a Targeted Attribute Perturbation Alignment Score (TAPAScore), which tests whether matched concepts respond selectively and in the expected direction under targeted image-level attribute perturbations. Under sanity checks, our matching and TAPAScore are the only evaluated metrics that reliably distinguish trained SAEs from untrained ones. Across SAEs trained on CLIP and DINOv2 embeddings, we find that increased overcompleteness can reduce perturbation alignment, indicating a reduction in interpretability. Our evaluation framework suggests that moderate dictionary sizes provide the best trade-off, yielding the most interpretable SAEs. Code and datasets are available at https://github.com/JonasKlotz/sae-concept-eval.

cs.CV↗

arXiv2026

Bribery's Influence on Ranked Aggregation

Kemeny Consensus is a well-known rank aggregation method in social choice theory. In this method, given a set of rankings, the goal is to find a ranking $\Pi$ that minimizes the total Kendall tau distance to the input rankings. Computing a Kemeny consensus is NP-hard, and even verifying whether a given ranking is a Kemeny consensus is coNP-complete. Fitzsimmons and Hemaspaandra [IJCAI 2021] established the computational intractability of achieving a desired consensus through manipulative actions. Kemeny Consensus is an optimisation problem related to Kemeny's rule. In this paper, we consider a decision problem related to Kemeny's rule, known as Kemeny Score, in which the goal is to decide whether there exists a ranking $\Pi$ whose total Kendall tau distance from the given rankings is at most $k$. Computation of Kemeny score is known to be NP-complete. In this paper, we investigate the impact of several manipulation actions on the Kemeny Score problem, in which given a set of rankings, an integer $k$, and a ranking $X$, the question is to decide whether it is possible to manipulate the given rankings so that the total Kendall tau distance of $X$ from the manipulated rankings is at most $k$. We show that this problem can be solved in polynomial time for various manipulation actions. Interestingly, these same manipulation actions are known to be computationally hard for Kemeny consensus.

cs.GT↗

arXiv2026

Data-Centric Benchmark for Label Noise Estimation and Ranking in Remote Sensing Binary Building Segmentation

High-quality pixel-level annotations are essential for the semantic segmentation of remote sensing imagery. However, such labels are expensive to obtain and often affected by noise due to the labor-intensive and time-consuming nature of pixel-wise annotation, which makes it challenging for human annotators to label every pixel accurately. Annotation errors can significantly degrade the performance and robustness of modern segmentation models, motivating the need for reliable mechanisms to identify and quantify noisy training samples. This paper introduces a novel data-centric benchmark, together with a new, publicly available binary building segmentation dataset. This specific task serves as a representative and practically relevant testbed that enables controlled experimentation with different annotation perturbations. Furthermore, we also introduce two techniques for identifying, quantifying, and ranking training samples according to their level of label noise in remote sensing semantic segmentation. Such proposed methods leverage complementary strategies based on model uncertainty, prediction consistency, and representation analysis, and consistently outperform established baselines across a range of experimental settings. The outcomes of this work are publicly available at https://github.com/keillernogueira/label_noise_segmentation.

cs.CV↗

arXiv2026

Robust Value Maximization in Challenge the Champ Tournaments with Probabilistic Outcomes

Challenge the Champ is a simple tournament format, where an ordering of the players -- called a seeding -- is decided. The first player in this order is the initial champ, and faces the next player. The outcome of each match decides the current champion, who faces the next player in the order. Each player also has a popularity, and the value of each match is the popularity of the winner. Value maximization in tournaments has been previously studied when each match has a deterministic outcome. However, match outcomes are often probabilistic, rather than deterministic. We study robust value maximization in Challenge the Champ tournaments, when the winner of a match may be probabilistic. That is, we seek to maximize the total value that is obtained, irrespective of the outcome of probabilistic matches. We show that even in simple binary settings, for non-adaptive algorithms, the optimal robust value -- which we term the \textsc{VnaR}, or the value not at risk -- is hard to approximate. However, if we allow adaptive algorithms that determine the order of challengers based on the outcomes of previous matches, or restrict the matches with probabilistic outcomes, we can obtain good approximations to the optimal \textsc{VnaR}.

cs.GT↗

arXiv2026

When agents choose bundles autonomously: guarantees beyond discrepancy

We consider the fair division of indivisible items among $n$ agents with additive non-negative normalized valuations, with the goal of obtaining high value guarantees, that is, close to the proportional share for each agent. We prove that partitions where \emph{every} part yields high value for each agent are asymptotically limited by a discrepancy barrier of $\Theta(\sqrt{n})$. Guided by this, our main objective is to overcome this barrier and achieve stronger individual guarantees for each agent in polynomial time. Towards this, we are able to exhibit an exponential improvement over the discrepancy barrier. In particular, we can create partitions on-the-go such that when agents arrive sequentially (representing a previously-agreed priority order) and pick a part autonomously and rationally (i.e., one of highest value), then each is guaranteed a part of value at least $\mathsf{PROP} - \mathcal{O}{(\log n)}$. Moreover, we show even better guarantees for three restricted valuation classes such as those defined by: a common ordering on items, a bound on the multiplicity of values, and a hypergraph with a bound on the \emph{influence} of any agent. Specifically, we study instances where: (1) the agents are ``close'' to unanimity in their relative valuation of the items -- a generalization of the ordered additive setting; (2) the valuation functions do not assign the same positive value to more than $t$ items; and (3) the valuation functions respect a hypergraph, a setting introduced by Christodoulou et al. [EC'23], where agents are vertices and items are hyperedges. While the sizes of the hyperedges and neighborhoods can be arbitrary, the influence of any agent $a$, defined as the number of its neighbors who value at least one item positively that $a$ also values positively, is bounded.

cs.GT↗

arXiv2026

Minimum Envy Graphical House Allocation Beyond Identical Valuations

House allocation is an extremely well-studied problem in the field of fair allocation, where the goal is to assign $n$ houses to $n$ agents while satisfying certain fairness criterion, e.g., envy-freeness. To model social interactions, the Graphical House Allocation framework introduces a social graph $G$, in which each vertex corresponds to an agent, and an edge $(u, v)$ corresponds to the potential of agent $u$ to envy the agent $v$, based on their allocations and valuations. In undirected social graphs, the potential for envy is in both the directions. In the Minimum Envy Graphical House Allocation (ME-GHA) problem, given a set of $n$ agents, $n$ houses, a social graph, and agent's valuation functions, the goal is to find an allocation that minimizes the total envy summed up over all the edges of $G$. Recent work, [Hosseini et al., AAMAS 2023, AAMAS 2024] studied ME-GHA in the regime of polynomial-time algorithms, and designed exact and approximation algorithms, for certain graph classes under identical agent valuations. We initiate the study of \gha with non-identical valuations, a setting that has so far remained unexplored. We investigate the multivariate (parameterized) complexity of \gha by identifying structural restrictions on the social graph and valuation functions that yield tractability. We also design moderately exponential-time algorithms for several graph classes, and a polynomial-time algorithm for {binary valuations that returns an allocation with envy at most one when the social graph has maximum degree at most one.

cs.GT↗

arXiv2025

TimeSenCLIP: A Time Series Vision-Language Model for Remote Sensing

Vision-language models (VLMs) have shown significant promise in remote sensing applications, particularly for land-use and land-cover (LULC) mapping via zero-shot classification and retrieval. However, current approaches face several key challenges, such as the dependence on caption-based supervision, which is often not available or very limited in terms of the covered semantics, and the fact of being adapted from generic VLM architectures that are suitable for very high resolution images. Consequently, these models tend to prioritize spatial context over spectral and temporal information, limiting their effectiveness for medium-resolution remote sensing imagery. In this work, we present TimeSenCLIP, a lightweight VLM for remote sensing time series, using a cross-view temporal contrastive framework to align multispectral Sentinel-2 time series with geo-tagged ground-level imagery, without requiring textual annotations. Unlike prior VLMs, TimeSenCLIP emphasizes temporal and spectral signals over spatial context, investigating whether single-pixel time series contain sufficient information for solving a variety of tasks.

cs.CV↗

arXiv2025

How to Resolve Envy by Adding Goods

We consider the problem of resolving the envy of a given initial allocation by adding elements from a pool of goods. We give a characterization of the instances where envy can be resolved by adding an arbitrary number of copies of the items in the pool. From this characterization, we derive a polynomial-time algorithm returning a respective solution if it exists. If the number of copies or the total number of added items are bounded, the problem becomes computationally intractable even in various restricted cases. We perform a parameterized complexity analysis, focusing on the number of agents and the pool size as parameters. Notably, although not every instance admits an envy-free solution, our approach allows us to efficiently determine, in polynomial time, whether a solution exists-an aspect that is both theoretically interesting and far from trivial.

cs.GT↗

arXiv2025

A Simple Algorithm for Combinatorial n-Fold ILPs Using the Steinitz Lemma

We present an algorithm for a class of $n$-fold ILPs whose existing algorithms in literature are often either (1) based on the \textit{augmentation framework} where one starts with an arbitrary solution and then iteratively moves towards an optimal solution by solving appropriate programs; or (2) require solving a linear relaxation of the program; or (3) are based on decomposition/proximity based arguments. Combinatorial $n$-fold ILPs is a class of $n$-fold ILPs introduced and studied by Knop et al. [MP2020] that captures several other problems in a variety of domains. We present a simple and direct algorithm that solves combinatorial $n$-fold ILPs with unbounded non-negative variables via an application of the Steinitz lemma. Depending on the structure of the input ILP, we also improve upon the existing algorithms in the literature in terms of the running time, thereby showing an improvement that mirrors the one shown by Rohwedder [ICALP2025] contemporaneously and independently.

cs.DS↗

arXiv2025

Efficient Algorithms for Electing Successive Committees

In a recently introduced model of successive committee elections (Bredereck et al., AAAI-20) for a given set of ordinal or approval preferences one aims to find a sequence of a given length of "best" same-size committees such that each candidate is a member of a limited number of consecutive committees. However, the practical usability of this model remains limited, as the described task turns out to be NP-hard for most selection criteria already for seeking committees of size three. Non-trivial or somewhat efficient algorithms for these cases are lacking too. Motivated by a desire to unlock the full potential of the described temporal model of committee elections, we devise (parameterized) algorithms that effectively solve the mentioned hard cases in realistic scenarios of a moderate number of candidates or of a limited time horizon.

cs.GT↗

arXiv2025

More Efforts Towards Fixed-Parameter Approximability of Multiwinner Rules

Multiwinner Elections have emerged as a prominent area of research with numerous practical applications. We contribute to this area by designing parameterized approximation algorithms and also resolving an open question by Yang and Wang [AAMAS'18]. More formally, given a set of candidates, \mathcal{C}, a set of voters,\mathcal{V}, approving a subset of candidates (called approval set of a voter), and an integer $k$, we consider the problem of selecting a ``good'' committee using Thiele rules. This problem is computationally challenging for most Thiele rules with monotone submodular satisfaction functions, as there is no (1-\frac{1}{e}-\epsilon)\footnote{Here, $e$ denotes the base of the natural logarithm.}-approximation algorithm in f(k)(|\mathcal{C}| + |\mathcal{V}|)^{o(k)} time for any fixed $\epsilon > 0$ and any computable function $f$, and no {\sf PTAS} even when the length of approval set is two. Skowron [WINE'16] designed an approximation scheme running in FPT time parameterized by the combined parameter, size of the approval set and $k$. In this paper, we consider a parameter $d+k$ (no $d$ voters approve the same set of $d$ candidates), where $d$ is upper bounded by the size of the approval set (thus, can be much smaller). With respect to this parameter, we design parameterized approximation schemes, a lossy polynomial-time preprocessing method, and show that an extra committee member suffices to achieve the desired score (i.e., $1$-additive approximation). Additionally, we resolve an open question by Yang and Wang~[AAMAS'18] regarding the fixed-parameter tractability of the problem under the PAV rule with the total score as the parameter, demonstrating that it admits an FPT algorithm.

cs.GT↗

arXiv2025

EcoWikiRS: Learning Ecological Representation of Satellite Images from Weak Supervision with Species Observations and Wikipedia

The presence of species provides key insights into the ecological properties of a location such as land cover, climatic conditions or even soil properties. We propose a method to predict such ecological properties directly from remote sensing (RS) images by aligning them with species habitat descriptions. We introduce the EcoWikiRS dataset, consisting of high-resolution aerial images, the corresponding geolocated species observations, and, for each species, the textual descriptions of their habitat from Wikipedia. EcoWikiRS offers a scalable way of supervision for RS vision language models (RS-VLMs) for ecology. This is a setting with weak and noisy supervision, where, for instance, some text may describe properties that are specific only to part of the species' niche or is irrelevant to a specific image. We tackle this by proposing WINCEL, a weighted version of the InfoNCE loss. We evaluate our model on the task of ecosystem zero-shot classification by following the habitat definitions from the European Nature Information System (EUNIS). Our results show that our approach helps in understanding RS images in a more ecologically meaningful manner. The code and the dataset are available at https://github.com/eceo-epfl/EcoWikiRS.

cs.CV↗

arXiv2025

Fairness and Efficiency in Two-Sided Matching Markets

We propose a new fairness notion, motivated by the practical challenge of allocating teaching assistants (TAs) to courses in a department. Each course requires a certain number of TAs and each TA has preferences over the courses they want to assist. Similarly, each course instructor has preferences over the TAs who applied for their course. We demand fairness and efficiency for both sides separately, giving rise to the following criteria: (i) every course gets the required number of TAs and the average utility of the assigned TAs meets a threshold; (ii) the allocation of courses to TAs is envy-free, where a TA envies another TA if the former prefers the latter's course and has a higher or equal grade in that course. Note that the definition of envy-freeness here differs from the one in the literature, and we call it merit-based envy-freeness. We show that the problem of finding a merit-based envy-free and efficient matching is NP-hard even for very restricted settings, such as two courses and uniform valuations; constant degree, constant capacity of TAs for every course, valuations in the range {0,1,2,3}, identical valuations from TAs, and even more. To find tractable results, we consider some restricted instances, such as, strict valuation of TAs for courses, the difference between the number of positively valued TAs for a course and the capacity, the number of positively valued TAs/courses, types of valuation functions, and obtained some polynomial-time solvable cases, showing the contrast with intractable results. We further studied the problem in the paradigm of parameterized algorithms and designed some exact and approximation algorithms.

cs.GT↗

arXiv2024

SenCLIP: Enhancing zero-shot land-use mapping for Sentinel-2 with ground-level prompting

Pre-trained vision-language models (VLMs), such as CLIP, demonstrate impressive zero-shot classification capabilities with free-form prompts and even show some generalization in specialized domains. However, their performance on satellite imagery is limited due to the underrepresentation of such data in their training sets, which predominantly consist of ground-level images. Existing prompting techniques for satellite imagery are often restricted to generic phrases like a satellite image of ..., limiting their effectiveness for zero-shot land-use and land-cover (LULC) mapping. To address these challenges, we introduce SenCLIP, which transfers CLIPs representation to Sentinel-2 imagery by leveraging a large dataset of Sentinel-2 images paired with geotagged ground-level photos from across Europe. We evaluate SenCLIP alongside other SOTA remote sensing VLMs on zero-shot LULC mapping tasks using the EuroSAT and BigEarthNet datasets with both aerial and ground-level prompting styles. Our approach, which aligns ground-level representations with satellite imagery, demonstrates significant improvements in classification accuracy across both prompt styles, opening new possibilities for applying free-form textual descriptions in zero-shot LULC mapping.

cs.CV↗

arXiv2024

When far is better: The Chamberlin-Courant approach to obnoxious committee selection

Classical work on metric space based committee selection problem interprets distance as ``near is better''. In this work, motivated by real-life situations, we interpret distance as ``far is better''. Formally stated, we initiate the study of ``obnoxious'' committee scoring rules when the voters' preferences are expressed via a metric space. To this end, we propose a model where large distances imply high satisfaction and study the egalitarian avatar of the well-known Chamberlin-Courant voting rule and some of its generalizations. For a given integer value $1 \le λ\le k$, the committee size k, a voter derives satisfaction from only the $λ$-th favorite committee member; the goal is to maximize the satisfaction of the least satisfied voter. For the special case of $λ= 1$, this yields the egalitarian Chamberlin-Courant rule. In this paper, we consider general metric space and the special case of a $d$-dimensional Euclidean space. We show that when $λ$ is $1$ and $k$, the problem is polynomial-time solvable in $\mathbb{R}^2$ and general metric space, respectively. However, for $λ= k-1$, it is NP-hard even in $\mathbb{R}^2$. Thus, we have ``double-dichotomy'' in $\mathbb{R}^2$ with respect to the value of λ, where the extreme cases are solvable in polynomial time but an intermediate case is NP-hard. Furthermore, this phenomenon appears to be ``tight'' for $\mathbb{R}^2$ because the problem is NP-hard for general metric space, even for $λ=1$. Consequently, we are motivated to explore the problem in the realm of (parameterized) approximation algorithms and obtain positive results. Interestingly, we note that this generalization of Chamberlin-Courant rules encodes practical constraints that are relevant to solutions for certain facility locations.

cs.DS↗

arXiv2024

Satisfiability to Coverage in Presence of Fairness, Matroid, and Global Constraints

In MaxSAT with Cardinality Constraint problem (CC-MaxSAT), we are given a CNF-formula $Φ$, and $k \ge 0$, and the goal is to find an assignment $β$ with at most $k$ variables set to true (also called a weight $k$-assignment) such that the number of clauses satisfied by $β$ is maximized. MaxCov can be seen as a special case of CC-MaxSAT, where the formula $Φ$ is monotone, i.e., does not contain any negative literals. CC-MaxSAT and MaxCov are extremely well-studied problems in the approximation algorithms as well as parameterized complexity literature. Our first contribution is that the two problems are equivalent to each other in the context of FPT-Approximation parameterized by $k$ (approximation is in terms of number of clauses satisfied/elements covered). We give a randomized reduction from CC-MaxSAT to MaxCov in time $O(1/ε)^{k} \cdot (m+n)^{O(1)}$ that preserves the approximation guarantee up to a factor of $1-ε$. Furthermore, this reduction also works in the presence of fairness and matroid constraints. Armed with this reduction, we focus on designing FPT-Approximation schemes (FPT-ASes) for MaxCov and its generalizations. Our algorithms are based on a novel combination of a variety of ideas, including a carefully designed probability distribution that exploits sparse coverage functions. These algorithms substantially generalize the results in Jain et al. [SODA 2023] for CC-MaxSAT and MaxCov for $K_{d,d}$-free set systems (i.e., no $d$ sets share $d$ elements), as well as a recent FPT-AS for Matroid-Constrained MaxCov by Sellier [ESA 2023] for frequency-$d$ set systems.

cs.DS↗

arXiv2024

Controlling Delegations in Liquid Democracy

In liquid democracy, agents can either vote directly or delegate their vote to a different agent of their choice. This results in a power structure in which certain agents possess more voting weight than others. As a result, it opens up certain possibilities of vote manipulation, including control and bribery, that do not exist in standard voting scenarios of direct democracy. Here we formalize a certain kind of election control -- in which an external agent may change certain delegation arcs -- and study the computational complexity of the corresponding combinatorial problem.

cs.GT↗