His research focuses on multi-modal learning, specifically advancing large vision-language models, embodied intelligence, and out-of-distribution detection. Xiang has published over 40 papers in top-tier venues, including CVPR, NeurIPS, ICML, AAAI, and ACM MM. He is the recipient of multiple awards, including the NTU Research Excellence Award and Best Student Paper at MIPR 2024, and serves as a reviewer for major AI conferences."

The most powerful artificial intelligence tools all have one thing in common. Today's most powerful AI tools - the ones that can summarise documents, generate artwork, write poetry or predict how incredibly complex proteins fold - all stand on the shoulders of the "transformer" . This neural network architecture, first announced in 2017 at an unassuming conference centre in California, enables machines to process information in a way that reflects how humans think. Previously, most state-of-the-art AI models relied on a technique called a recurrent neural network . This worked by reading text in tight windows, left to right, remembering only what came just before.

Lego also recognized that it had to build a course that'll work regardless of a teacher's fluency in such subjects. So a big part of developing the course was making sure that teachers had the tools they needed to be on top of whatever lessons they're working on. "When we design and we test the products, we're not the ones testing in the classroom," Silwinski said. "We give it to a teacher and we provide all of the lesson materials, all of the training, all of the notes, all the presentation materials, everything that they need to be able to teach the lesson." Lego also took into account the fact that some schools might introduce its students to these things starting in Kindergarten, whereas others might skip to the grade 3-5 or 6-8 sets.

Amateur mathematicians are using artificial intelligence chatbots to solve long-standing problems, in a move that has taken professionals by surprise. While the problems in question aren't the most advanced in the mathematical canon, the success of AI models in tackling them shows that their mathematical performance has passed a significant threshold, say researchers, and could fundamentally change the way we do mathematics. The questions being solved by AI originate from Hungarian mathematician Paul Erdős, who was famous for his ability to pose useful but difficult questions during a career that spanned over six decades. "The questions tended to be very simple, but very hard," says Thomas Bloom at the University of Manchester, UK. By his death in 1996, there were more than 1000 of these unsolved Erdős problems, spanning a wide range of mathematical disciplines, from combinatorics (the study of combinations) to number theory.

If your New Year's resolution is to understand quantum computing this year, take a cue from a 9-year-old podcaster talking to some of the biggest minds in the field, says quantum columnist Karmela Padavic-Callaghan Quantum computing seems to pop up in the news pretty often these days. You've probably seen quantum chips gracing your feeds and their odd, steampunk-ish cooling systems in the pages of magazines and newspapers. Politicians and business leaders are peppering their announcements with the word "quantum" more frequently, too. If you're feeling a little confused about it all, it's a good year for a New Year's resolution to finally figure out what quantum computing is all about. This is an ambitious goal, and the timing certainly makes sense.

This material may not be published, broadcast, rewritten, or redistributed. Quotes displayed in real-time or delayed by at least 15 minutes. Market data provided by Factset . Powered and implemented by FactSet Digital Solutions . Mutual Fund and ETF data provided by LSEG .

Can ChatGPT really help you change your life - or just flatter you? Can ChatGPT really help you change your life - or just flatter you? AI as a life coach: experts share what works, what doesn't and what to look out for It's becoming more common for people to use AI chatbots for personal guidance - but this doesn't come without risks Setting goals is hard; keeping them is harder - and failure can bring about icky feelings about yourself. This year, in an effort to game the system and tilt the scales toward success, some people used AI for their 2026 resolutions. It's the latest step in an ongoing trend: in September 2025, OpenAI, the company behind ChatGPT, released findings showing that using the AI chatbot for personal guidance is very common.

We (AIhub) will be running a short course on science communication on Wednesday 21 January, from 13:00 - 14:30. In this brief tutorial, science communication experts will teach you how to clearly and concisely explain your research to non-specialists.

In this paper, we extend the Riesz representation framework to causal inference under sample selection, where both treatment assignment and outcome observability are non-random. Formulating the problem in terms of a Riesz representer enables stable estimation and a transparent decomposition of omitted variable bias into three interpretable components: a data-identified scale factor, outcome confounding strength, and selection confounding strength. For estimation, we employ the ForestRiesz estimator, which accounts for selective outcome observability while avoiding the instability associated with direct propensity score inversion. We assess finite-sample performance through a simulation study and show that conventional double machine learning approaches can be highly sensitive to tuning parameters due to their reliance on inverse probability weighting, whereas the ForestRiesz estimator delivers more stable performance by leveraging automatic debiased machine learning. In an empirical application to the gender wage gap in the U.S., we find that our ForestRiesz approach yields larger treatment effect estimates than a standard double machine learning approach, suggesting that ignoring sample selection leads to an underestimation of the gender wage gap. Sensitivity analysis indicates that implausibly strong unobserved confounding would be required to overturn our results. Overall, our approach provides a unified, robust, and computationally attractive framework for causal inference under sample selection.

Deep double descent is one of the key phenomena underlying the generalization capability of deep learning models. In this study, epoch-wise double descent, which is delayed generalization following overfitting, was empirically investigated by focusing on the evolution of internal structures. Fully connected neural networks of three different sizes were trained on the CIFAR-10 dataset with 30% label noise. By decomposing the loss curves into signal contributions from clean and noisy training data, the epoch-wise evolutions of internal signals were analyzed separately. Three main findings were obtained from this analysis. First, the model achieved strong re-generalization on test data even after perfectly fitting noisy training data during the double descent phase, corresponding to a "benign overfitting" state. Second, noisy data were learned after clean data, and as learning progressed, their corresponding internal activations became increasingly separated in outer layers; this enabled the model to overfit only noisy data. Third, a single, very large activation emerged in the shallow layer across all models; this phenomenon is referred as "outliers," "massive activations," and "super activations" in recent large language models and evolves with re-generalization. These empirical findings directly link the recent key phenomena of "deep double descent," "benign over-fitting," and "large activation", and support the proposal of a novel scenario for understanding deep double descent. Artificial intelligence technologies have undergone remarkable development in recent years, introducing substantial transformation to social structures and influencing various academic fields. Although these models form the core of such technologies, the fundamental principles underlying their high generalization capability when trained on real-world data remain poorly understood. Recent numerical experiments have empirically revealed various intriguing phenomena related to this gap.