arXiv subjects

Dimitris Paparas

Publications and source records attributed to Dimitris Paparas.

11 recordsLinked to original sources

arXiv2026

TCS-BENCH: Benchmarking State-of-the-Art Generative AI Theoretical Computer Science Research Ability

We introduce TCS-Bench, a benchmark for evaluating Large Language Models (LLMs) on research-level Theoretical Computer Science (TCS) proof generation. TCS-Bench consists of theorem-proving tasks from papers published at top theoretical computer science venues (STOC, FOCS, and SODA). Each task provides the necessary context to derive a self-contained proof for a target result. We evaluate state-of-the-art models on this benchmark. We verify the correctness of generated proofs via a verification agent, and further benchmark the verifier against human-expert proof judgements on a set of target statements and generated proofs pairs. Our reference verifier achieves over 90% accuracy on the expert labeled set.

cs.CL↗

arXiv2026

The Magic Correlations: Understanding Knowledge Transfer from Pretraining to Supervised Fine-Tuning

Understanding how language model capabilities transfer from pretraining to supervised fine-tuning (SFT) is fundamental to efficient model development and data curation. In this work, we investigate four core questions: RQ1. To what extent do accuracy and confidence rankings established during pretraining persist after SFT? RQ2. Which benchmarks serve as robust cross-stage predictors and which are unreliable? RQ3. How do transfer dynamics shift with model scale? RQ4. How well does model confidence align with accuracy, as a measure of calibration quality? Does this alignment pattern transfer across training stages? We address these questions through a suite of correlation protocols applied to accuracy and confidence metrics across diverse data mixtures and model scales. Our experiments reveal that transfer reliability varies dramatically across capability categories, benchmarks, and scales -- with accuracy and confidence exhibiting distinct, sometimes opposing, scaling dynamics. These findings shed light on the complex interplay between pretraining decisions and downstream outcomes, providing actionable guidance for benchmark selection, data curation, and efficient model development.

cs.LG↗

arXiv2026

Accelerating Scientific Research with Gemini: Case Studies and Common Techniques

Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to contribute to novel, expert-level mathematical discovery is less understood. We present a collection of case studies demonstrating how researchers have successfully collaborated with advanced AI models, specifically Google's Gemini-based models (in particular Gemini Deep Think and its advanced variants), to solve open problems, refute conjectures, and generate new proofs across diverse areas in theoretical computer science, as well as other areas such as economics, optimization, and physics. Based on these experiences, we extract common techniques for effective human-AI collaboration in theoretical research, such as iterative refinement, problem decomposition, and cross-disciplinary knowledge transfer. While the majority of our results stem from this interactive, conversational methodology, we also highlight specific instances that push beyond standard chat interfaces. These include deploying the model as a rigorous adversarial reviewer to detect subtle flaws in existing proofs, and embedding it within a "neuro-symbolic" loop that autonomously writes and executes code to verify complex derivations. Together, these examples highlight the potential of AI not just as a tool for automation, but as a versatile, genuine partner in the creative process of scientific discovery.

cs.CL↗

arXiv2026

The Query Complexity of Local Search in Rounds on General Graphs

We analyze the query complexity of finding a local minimum in $t$ rounds on general graphs. More precisely, given a graph $G = (V,E)$ and oracle access to an unknown function $f : V \to \mathbb{R}$, the goal is to find a local minimum--a vertex $v$ such that $f(v) \leq f(u)$ for all $(u,v) \in E$--using at most $t$ rounds of interaction with the oracle. The query complexity is well understood on grids, but much less is known beyond. This abstract problem captures many optimization tasks, such as finding a local minimum of a loss function during neural network training. For each graph with $n$ vertices, we prove a deterministic upper bound of $O(t n^{1/t} (s\Delta)^{1-1/t})$, where $s$ is the separation number and $\Delta$ is the maximum degree of the graph. We complement this result with a randomized lower bound of $\Omega(t n^{1/t}-t)$ that holds for any connected graph. We also find that parallel steepest descent with a warm start provides improved bounds for graphs with high separation number and bounded degree. To obtain our results, we utilized an advanced version of Gemini at various stages of our research. We discuss our experience in a methodology section.

cs.CC↗

arXiv2025

Gemma 3 Technical Report

We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achieved by increasing the ratio of local to global attention layers, and keeping the span on local attention short. The Gemma 3 models are trained with distillation and achieve superior performance to Gemma 2 for both pre-trained and instruction finetuned versions. In particular, our novel post-training recipe significantly improves the math, chat, instruction-following and multilingual abilities, making Gemma3-4B-IT competitive with Gemma2-27B-IT and Gemma3-27B-IT comparable to Gemini-1.5-Pro across benchmarks. We release all our models to the community.

cs.CL↗

arXiv2024

Kernel Banzhaf: A Fast and Robust Estimator for Banzhaf Values

Banzhaf values provide a popular, interpretable alternative to the widely-used Shapley values for quantifying the importance of features in machine learning models. Like Shapley values, computing Banzhaf values exactly requires time exponential in the number of features, necessitating the use of efficient estimators. Existing estimators, however, are limited to Monte Carlo sampling methods. In this work, we introduce Kernel Banzhaf, the first regression-based estimator for Banzhaf values. Our approach leverages a novel regression formulation, whose exact solution corresponds to the exact Banzhaf values. Inspired by the success of Kernel SHAP for Shapley values, Kernel Banzhaf efficiently solves a sampled instance of this regression problem. Through empirical evaluations across eight datasets, we find that Kernel Banzhaf significantly outperforms existing Monte Carlo methods in terms of accuracy, sample efficiency, robustness to noise, and feature ranking recovery. Finally, we complement our experimental evaluation with strong theoretical guarantees on Kernel Banzhaf's performance.

cs.LG↗

arXiv2024

Scaling Laws for Downstream Task Performance of Large Language Models

Scaling laws provide important insights that can guide the design of large language models (LLMs). Existing work has primarily focused on studying scaling laws for pretraining (upstream) loss. However, in transfer learning settings, in which LLMs are pretrained on an unsupervised dataset and then finetuned on a downstream task, we often also care about the downstream performance. In this work, we study the scaling behavior in a transfer learning setting, where LLMs are finetuned for machine translation tasks. Specifically, we investigate how the choice of the pretraining data and its size affect downstream performance (translation quality) as judged by: downstream cross-entropy and translation quality metrics such as BLEU and COMET scores. Our experiments indicate that the size of the finetuning dataset and the distribution alignment between the pretraining and downstream data significantly influence the scaling behavior. With sufficient alignment, both downstream cross-entropy and translation quality scores improve monotonically with more pretraining data. In such cases, we show that it is possible to predict the downstream translation quality metrics with good accuracy using a log-law. However, there are cases where moderate misalignment causes the downstream translation scores to fluctuate or get worse with more pretraining, whereas downstream cross-entropy monotonically improves. By analyzing these, we provide new practical insights for choosing appropriate pretraining data.

cs.CL↗

arXiv2020

Searching, Sorting, and Cake Cutting in Rounds

We study searching and sorting in rounds motivated by a fair division question: given a cake cutting problem with $n$ players, compute a fair allocation in at most $k$ rounds of interaction with the players. Rounds interpolate between the simultaneous and the fully adaptive settings, also capturing parallel complexity. We find that proportional cake cutting in rounds is equivalent to sorting with rank queries in rounds. We design a protocol for proportional cake cutting in rounds, while lower bounds for sorting in rounds with rank queries were given by Alon and Azar. Inspired by the rank query model, we then consider two basic search problems: ordered and unordered search. In unordered search, we get an array $\vec{x}=(x_1, \ldots, x_n)$ and an element $z$ promised to be in $\vec{x}$. We have access to an oracle that receives queries of the form "Is $z$ at location $i$?" and answers "Yes" or "No". The goal is to find the location of $z$ with success probability at least $p$ in at most $k$ rounds of interaction with the oracle. We show the expected query complexity of randomized algorithms on a worst case input is $np\bigl(\frac{k+1}{2k}\bigr) \pm O(1)$, while that of deterministic algorithms on a worst case input distribution is $np \bigl(1 - \frac{k-1}{2k}p \bigr) \pm O(1)$. These bounds apply even to fully adaptive unordered search, where the ratio between the two complexities converges to $2-p$ as the size of the array grows. In ordered search, we get sorted array $\vec{x}=(x_1, \ldots, x_n)$ and element $z$ promised to be in $\vec{x}$. We have access to an oracle that gets comparison queries. Here we find that the expected query complexity of randomized algorithms on a worst case input and deterministic algorithms on a worst case input distribution is essentially the same: $p k \cdot n^{\frac{1}{k}} \pm O(1+pk)$.

cs.DS↗

arXiv2017

On the Complexity of Simple and Optimal Deterministic Mechanisms for an Additive Buyer

We show that the Revenue-Optimal Deterministic Mechanism Design problem for a single additive buyer is #P-hard, even when the distributions have support size 2 for each item and, more importantly, even when the optimal solution is guaranteed to be of a very simple kind: the seller picks a price for each individual item and a price for the grand bundle of all the items; the buyer can purchase either the grand bundle at its given price or any subset of items at their total individual prices. The following problems are also #P-hard, as immediate corollaries of the proof: 1. determining if individual item pricing is optimal for a given instance, 2. determining if grand bundle pricing is optimal, and 3. computing the optimal (deterministic) revenue. On the positive side, we show that when the distributions are i.i.d. with support size 2, the optimal revenue obtainable by any mechanism, even a randomized one, can be achieved by a simple solution of the above kind (individual item pricing with a discounted price for the grand bundle) and furthermore, it can be computed in polynomial time. The problem can be solved in polynomial time too when the number of items is constant.

cs.GT↗

arXiv2013

The Complexity of Optimal Multidimensional Pricing

We resolve the complexity of revenue-optimal deterministic auctions in the unit-demand single-buyer Bayesian setting, i.e., the optimal item pricing problem, when the buyer's values for the items are independent. We show that the problem of computing a revenue-optimal pricing can be solved in polynomial time for distributions of support size 2, and its decision version is NP-complete for distributions of support size 3. We also show that the problem remains NP-complete for the case of identical distributions.

cs.GT↗

arXiv2012

The Complexity of Non-Monotone Markets

We introduce the notion of non-monotone utilities, which covers a wide variety of utility functions in economic theory. We then prove that it is PPAD-hard to compute an approximate Arrow-Debreu market equilibrium in markets with linear and non-monotone utilities. Building on this result, we settle the long-standing open problem regarding the computation of an approximate Arrow-Debreu market equilibrium in markets with CES utility functions, by proving that it is PPAD-complete when the Constant Elasticity of Substitution parameter ρis any constant less than -1.

cs.CC↗