AI4X 2025 Papers

Financial markets, characterized by their volatility, uncertainty, complexity, and ambiguity (VUCA), pose significant challenges for accurate predictions. Investment has become increasingly intertwined with technological advancements, as machine learning models revolutionise the field of stock market trend predictions, offering potential solutions by processing large datasets, identifying trends, and minimizing human bias. While machine learning is increasingly applied in financial forecasting, understanding the relative strengths and weaknesses of different algorithms across varying time frames remains underexplored. This is especially relevant given the rise of algorithmic trading and new stock markets such as cryptocurrencies, underscoring the need for precise, data-driven predictions. The aim of this study is to evaluate the performance of machine learning algorithms in predicting stock prices within Singapore’s banking sector. The study explores how each algorithm performs when trained on different amounts of data, comparing its effectiveness for short-term, mid-term and long-term stock price predictions. To do this, historical stock prices were collected using the Yahoo Finance API, focusing on closing prices as the target variable. Using the data collected from major Singaporean banks, namely DBS, OCBC and UOB, this study evaluated the performance of various machine learning algorithms: Random Forest (RF), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM). The various models were all trained on different datasets, and its predictions for the closing price on a specific date was recorded. Each model was evaluated using rigorous performance metrics, including percentage error, R² values, mean absolute error, and mean squared error, to determine their efficacy in capturing trends and minimising predictive inaccuracies. Different algorithms have distinct methods of learning patterns and handling data variability, and thus will perform differently under the same conditions. Hence, we hoped to gain greater insight into each model’s performance and assess their adaptability to the various time frames. This study contributes to the growing body of research on AI-driven financial forecasting by providing a comparative analysis of machine learning algorithms in Singapore’s banking sector. It highlights the need for flexibility in one’s approach to algorithmic trading to enhance prediction accuracy across diverse scenarios. The insights gained can aid financial analysts, traders, and decision-makers in developing data-driven strategies for stock market investments, ultimately promoting more informed decision-making and risk management in a volatile financial landscape.

By integrating artificial intelligence into the material search process, we gain new insights into the correlation between crystallographic parameters and optical anisotropy of van der Waals (vdW) materials, which was confirmed experimentally.

Abstract Metal–organic frameworks (MOFs) are highly porous and versatile materials studied extensively for applications such as carbon capture and water harvesting. However, computing phonon-mediated properties in MOFs, like thermal expansion and mechanical stability, remains challenging due to the large number of atoms per unit cell, making traditional Density Functional Theory (DFT) methods impractical for high-throughput screening. Recent advances in machine learning potentials have led to foundation atomistic models, such as MACE-MP-0, that accurately predict equilibrium structures but struggle with phonon properties of MOFs. In this work, we developed a workflow for computing phonons in MOFs within the quasi-harmonic approximation with a fine-tuned MACE model, MACE-MP-MOF0. The model was trained on a curated dataset of 127 representative and diverse MOFs. The fine-tuned MACE-MP-MOF0 improves the accuracy of phonon density of states and corrects the imaginary phonon modes of MACE-MP-0, enabling high-throughput phonon calculations with state-of-the-art precision. The model successfully predicts thermal expansion and bulk moduli in agreement with DFT and experimental data for several well-known MOFs. These results highlight the potential of MACE-MP-MOF0 in guiding MOF design for applications in energy storage and thermoelectrics.

Accurately predicting Alzheimer's Disease (AD) progression is useful for clinical care. The 2019 TADPOLE (The Alzheimer's Disease Prediction Of Longitudinal Evolution) challenge evaluated 92 algorithms from 33 teams worldwide. Unlike typical clinical prediction studies, TADPOLE accommodates (1) a variable number of observed time points across patients, (2) missing data across modalities and visits, and (3) prediction over an open-ended time horizon, which better reflects real-world data. However, TADPOLE only used the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, so how well top algorithms generalize to other cohorts remains unclear. We tested five algorithms in three external datasets covering 2312 participants and 13,200 time points. The algorithms included FROG, the overall TADPOLE winner, which utilized a unique Longitudinal-to-Cross-sectional (L2C) transformation to convert variable-length longitudinal histories into feature vectors of the same length across participants (i.e., same-length feature vectors). We also considered two FROG variants. One variant unified all XGBoost models from the original FROG with a single feedforward neural network (FNN), which we referred to as L2C-FNN. We also included minimal recurrent neural networks (MinimalRNN), which were ranked second at publication time, as well as AD Course Map (AD-Map), which outperformed MinimalRNN at publication time. All five models-three FROG variants, MinimalRNN, and AD-Map-were trained on ADNI and tested on the external datasets. L2C-FNN performed the best overall. In the case of predicting cognition and ventricle volume, L2C-FNN and AD-Map were the best. For clinical diagnosis prediction, L2C-FNN was the best, while AD-Map was the worst. L2C-FNN also maintained its edge over other models, regardless of the number of observed time points and regardless of the prediction horizon from 0 to 6 years into the future. L2C-FNN shows strong potential for both short-term and long-term dementia progression prediction. Pretrained ADNI models are available: https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/predict_phenotypes/Zhang2025_L2CFNN.

In this paper, we study the utility-cost ratio (UCR) maximization problem in quantum networks, with a particular focus on secure quantum key distribution (QKD). We propose the Quantum Utility-Cost Optimization (QUCO) algorithm to address the challenge of balancing network utility with operational and infrastructure costs. The utility of the network is measured based on the secret key fraction, a key performance metric in QKD, while the costs considered include operational and maintenance expenses, transceiver deployment, auxiliary equipment, and wavelength channel usage. The QUCO algorithm transforms the original non-convex problem into a convex optimization through a series of transformations, enabling efficient resource allocation across quantum network routes. Our simulation results, conducted on a realistic quantum network topology, demonstrate the algorithm’s ability to improve the utility-cost ratio while maintaining high network performance. The QUCO algorithm strikes a balance between minimizing costs and maintaining high end-to-end entanglement fidelity, which is effective for optimizing UCR in quantum networks.