Paraskevas Lekeas

Dr. Paraskevas Lekeas

orcid

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 January 2013 in SSRN Electronic Journal
Verified
Preprint
Open access
PDF file
PLParaskevas Lekeas
Paraskevas Lekeas
GSGiorgos Stamatopoulos
Giorgos Stamatopoulos
We analyze cooperative Cournot games with boundedly rational firms. Due to cognitive constraints, the members of a coalition cannot accurately predict the coali

20 February 2012 in arXiv
Verified
Preprint
Open access
PDF file
PLParaskevas Lekeas
Paraskevas Lekeas
In Network cooperative games, due to computational complexity issues, agents are not able to base their behavior on the "whole network status" but have to follow certain "beliefs" as to how it is in their strategic interest to act. This behavior constitutes the main interest of this paper. To this end, we quantify and characterize the set of beliefs that support cooperation of such agents. Assuming that they are engaged in a differentiated Cournot competition and that they equally split the worth produced, we characterize the set of coalitional beliefs that support core non-emptiness and thus guarantee stability of the Network.

29 July 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
PLParaskevas Lekeas
Paraskevas Lekeas
GSGiorgos Stamatopoulos
Giorgos Stamatopoulos
LLM agents are known to deviate from Nash equilibria in strategic interactions, but nobody has looked inside the model to understand why, or asked whether the deviation can be reversed. We do both. Working with four open-source models (Llama-3 and Qwen2.5, 8B to 72B parameters) playing four canonical two-player games, we first establish the behavioral picture through self-play and cross-play experiments, then open up the 32-layer Llama-3-8B model and examine what actually happens during a strategic decision. The mechanistic findings are clear. Opponent history is encoded with near-perfect fidelity at the very first layer (96% probe accuracy) and consumed progressively by later ones, while Nash action encoding is weak throughout, never exceeding 56%. There is no dedicated Nash module. Instead, the model privately favors the Nash action through most of its forward pass, but a prosocial override (a bias toward cooperative, other-regarding behavior rooted in pretraining on human text and further modulated by RLHF) concentrated in the final layers reverses this, reaching 84% probability of cooperation at layer 30. When we inject a learned Nash direction into the residual stream, the behavior shifts bidirectionally and causally, confirmed through concept clamping. The behavioral experiments surface six scale-and architecture-dependent findings in self-play, the most notable being that chain-of-thought reasoning worsens Nash play in small models but achieves near-perfect Nash play in models above 70B parameters. The crossplay experiments reveal three phenomena invisible in self-play: a small model can unravel the cooperation of any partner simply by defecting early; two large models reinforce each other's cooperative instincts indefinitely; and who moves first in a coordination game determines which Nash equilibrium the system lands on. The central finding is that LLMs do not lack Nashplaying competence. They compute it, then suppress it.

1 January 2013 in Lecture Notes in Computer Science
Verified
Book Chapter
Closed access
PDF file
PLParaskevas Lekeas
Paraskevas Lekeas

18 April 2012 in arXiv
Verified
Preprint
Open access
PDF file
PLParaskevas Lekeas
Paraskevas Lekeas
Existing web infrastructures have supported the publication of a tremendous amount of resources, and over the past few years Data Resource Usage is an everyday task for millions of users all over the world. In this work we model Resource Usage as a Cooperative Cournot Game in which a resource user and the various resource services are engaged. We give quantified answers as to when it is of interest for the user to stop using part of a resource and to switch to a different one. Moreover, we do the same from the perspective of a resource's provider.

4 December 2013 in arXiv
Verified
Preprint
Open access
PDF file
PLParaskevas Lekeas
Paraskevas Lekeas
In Raman (2007), using a column construction technique it is proved that every interval graph can be colored online with First Fit with at most $8w(G)$ colors, where $w(G)$ is the size of the maximum clique of $G$. Since the column construction can not be adapted to circular arc graphs we give a different proof to establish an upper bound of $9w(G)$ for online coloring a circular arc graph $G$ with the First Fit algorithm.