Searcharxiv⌕ Search

arXiv subjects

Yosuke Kikuchi

Publications and source records attributed to Yosuke Kikuchi.

2 recordsLinked to original sources

Molecular notation can act as a shortcut in AI benchmark evaluation

High predictive accuracy does not necessarily mean that an AI model relies on information relevant to its scientific task. Molecular prediction illustrates this concern: the same molecule can be represented by different SMILES strings, potentially altering model behavior. We combined a systematic PubMed literature survey with experiments on chemical language models. SMILES preparation could not be identified from articles or Supplementary Information in 71 of 174 studies (40.8%). Across six benchmarks, 58.7% of identity-preserving input pairs used different SMILES strings. Notation differences changed reconstruction and latent representations, while separately trained classifiers often achieved similar scores. In two datasets, notation was associated with labels and changed predictions of a fixed model, showing how benchmark scores can reward notational shortcuts. Balancing notation during classifier training reduced prediction differences but sometimes lowered performance on original inputs. These findings support evaluating predictive performance alongside consistency under identity-preserving changes in input representation.

q-bio.QM↗

A quantum protocol to win the graph colouring game on all Hadamard graphs

This paper deals with graph colouring games, an example of pseudo-telepathy, in which two provers can convince a verifier that a graph $G$ is $c$-colourable where $c$ is less than the chromatic number of the graph. They win the game if they convince the verifier. It is known that the players cannot win if they share only classical information, but they can win in some cases by sharing entanglement. The smallest known graph where the players win in the quantum setting, but not in the classical setting, was found by Galliard, Tapp and Wolf and has 32,768 vertices. It is a connected component of the Hadamard graph $G_N$ with $N=c=16$. Their protocol applies only to Hadamard graphs where $N$ is a power of 2. We propose a protocol that applies to all Hadamard graphs. Combined with a result of Frankl, this shows that the players can win on any induced subgraph of $G_{12}$ having 1609 vertices, with $c=12$. Combined with a result of Frankl and Rodl, our result shows that all sufficiently large Hadamard graphs yield pseudo-telepathy games.

quant-ph↗