arXiv subjects

Hideyuki Takada

Publications and source records attributed to Hideyuki Takada.

6 recordsLinked to original sources

arXiv2025

Improving Translation Quality by Selecting Better Data for LLM Fine-Tuning: A Comparative Analysis

We investigated the impact of data selection on machine translation fine-tuning for open LLMs. Using Japanese-English corpora, we compare five selectors: TF-IDF, COMET Kiwi, QuRate, FD-Score, and random selection, under controlled training conditions. We observed that semantic selectors consistently outperform lexical and geometry-based heuristics, and that even when the selected data differ by less than 3%, the impact on model performance is substantial, underscoring the sensitivity of fine-tuning to data quality.

cs.CL↗

arXiv2025

Exploring Parameter-Efficient Fine-Tuning and Backtranslation for the WMT 25 General Translation Task

In this paper, we explore the effectiveness of combining fine-tuning and backtranslation on a small Japanese corpus for neural machine translation. Starting from a baseline English{\textrightarrow}Japanese model (COMET = 0.460), we first apply backtranslation (BT) using synthetic data generated from monolingual Japanese corpora, yielding a modest increase (COMET = 0.468). Next, we fine-tune (FT) the model on a genuine small parallel dataset drawn from diverse Japanese news and literary corpora, achieving a substantial jump to COMET = 0.589 when using Mistral 7B. Finally, we integrate both backtranslation and fine-tuning{ -- }first augmenting the small dataset with BT generated examples, then adapting via FT{ -- }which further boosts performance to COMET = 0.597. These results demonstrate that, even with limited training data, the synergistic use of backtranslation and targeted fine-tuning on Japanese corpora can significantly enhance translation quality, outperforming each technique in isolation. This approach offers a lightweight yet powerful strategy for improving low-resource language pairs.

cs.CL↗

arXiv2021

The Black-Scholes Equation in Presence of Arbitrage

We apply Geometric Arbitrage Theory to obtain results in Mathematical Finance, which do not need stochastic differential geometry in their formulation. First, for a generic market dynamics given by a multidimensional Itô's process we specify and prove the equivalence between (NFLVR) and expected utility maximization. As a by-product we provide a geometric characterization of the (NUPBR) condition given by the zero curvature (ZC) condition. Finally, we extend the Black-Scholes PDE to markets allowing arbitrage.

q-fin.PR↗

arXiv2021

Can You hear the Shape of a Market? Geometric Arbitrage and Spectral Theory

Geometric Arbitrage Theory reformulates a generic asset model possibly allowing for arbitrage by packaging all assets and their forwards dynamics into a stochastic principal fibre bundle, with a connection whose parallel transport encodes discounting and portfolio rebalancing, and whose curvature measures, in this geometric language, the 'instantaneous arbitrage capability' generated by the market itself. The cashflow bundle is the vector bundle associated to this stochastic principal fibre bundle for the natural choice of the vector space fibre. The cashflow bundle carries a stochastic covariant differentiation induced by the connection on the principal fibre bundle. The link between arbitrage theory and spectral theory of the connection Laplacian on the vector bundle is given by the zero eigenspace resulting in a parametrization of all risk neutral measures equivalent to the statistical one. This indicates that a market satisfies the (NFLVR) condition if and only if $0$ is in the discrete spectrum of the connection Laplacian on the cash flow bundle or of the Dirac Laplacian of the twisted cash flow bundle with the exterior algebra bundle. We apply this result by extending Jarrow-Protter-Shimbo theory of asset bubbles for complete arbitrage free markets to markets not satisfying the (NFLVR). Moreover, by means of the Atiyah-Singer index theorem, we prove that the Euler characteristic of the asset nominal space is a topological obstruction to the the (NFLVR) condition, and, by means of the Bochner-Weitzenböck formula, the non vanishing of the homology group of the cash flow bundle is revealed to be a topological obstruction to (NFLVR), too. Asset bubbles are defined, classified and decomposed for markets allowing arbitrage.

q-fin.MF↗

arXiv2021

Credit Bubbles in Arbitrage Markets: The Geometric Arbitrage Approach to Credit Risk

We apply Geometric Arbitrage Theory to obtain results in mathematical finance for credit markets, which do not need stochastic differential geometry in their formulation. We obtain closed form equations involving default intensities and loss given defaults characterizing the no-free-lunch-with-vanishing-risk condition for corporate bonds, as well as the generic dynamics for credit market allowing for arbitrage possibilities. Moreover, arbitrage credit bubbles for both base credit assets and credit derivatives are explicitly computed for the market dynamics minimizing the arbitrage.

q-fin.PR↗

arXiv2021

When Risks and Uncertainties Collide: Mathematical Finance for Arbitrage Markets in a Quantum Mechanical View

Geometric arbitrage theory reformulates a generic asset model possibly allowing for arbitrage by packaging all asset and their forward dynamics into a stochastic principal fibre bundle, with a connection whose parallel transport encodes discounting and portfolio rebalancing, and whose curvature measures, in this geometric language, the instantaneous arbitrage capability generated by the market itself. The asset and market portfolio dynamics have a quantum mechanical description, which is constructed by quantizing the deterministic version of the stochastic Lagrangian system describing a market allowing for arbitrage. Results, obtained by solving the Schroedinger equation, coincide with those obtained by solving the stochastic Euler Lagrange equations derived by a variational principle and providing therefore consistency.

q-fin.RM↗