Despite recent algorithmic advances, we still lack principled ways to leverage the well-documented rescaling symmetries in ReLU neural network parameters. While two properly rescaled weights implement the same function, the training dynamics can be dramatically different. To offer a fresh perspective on exploiting this phenomenon, we build on the recent path-lifting framework, which provides a compact factorization of ReLU networks. We introduce a geometrically motivated criterion to rescale neural network parameters which minimization leads to a conditioning strategy that aligns a kernel in the path-lifting space with a chosen reference. We derive an efficient algorithm to perform this alignment. In the context of random network initialization, we analyze how the architecture and the initialization scale jointly impact the output of the proposed method. Numerical experiments illustrate its potential to speed up training.

We study post-calibration uncertainty for trained ensembles of classifiers. Specifically, we consider both aleatoric (label noise) and epistemic (model) uncertainty. Among the most popular and widely used calibration methods in classification are temperature scaling (i.e., pool-then-calibrate) and conformal methods. However, the main shortcoming of these calibration methods is that they do not balance the proportion of aleatoric and epistemic uncertainty. Not balancing these uncertainties can severely misrepresent predictive uncertainty, leading to overconfident predictions in some input regions while being underconfident in others. To address this shortcoming, we present a simple but powerful calibration algorithm Joint Uncertainty Calibration (JUCAL) that jointly calibrates aleatoric and epistemic uncertainty. JUCAL jointly calibrates two constants to weight and scale epistemic and aleatoric uncertainties by optimizing the negative log-likelihood (NLL) on the validation/calibration dataset. JUCAL can be applied to any trained ensemble of classifiers (e.g., transformers, CNNs, or tree-based methods), with minimal computational overhead, without requiring access to the models' internal parameters. We experimentally evaluate JUCAL on various text classification tasks, for ensembles of varying sizes and with different ensembling strategies. Our experiments show that JUCAL significantly outperforms SOTA calibration methods across all considered classification tasks, reducing NLL and predictive set size by up to 15% and 20%, respectively. Interestingly, even applying JUCAL to an ensemble of size 5 can outperform temperature-scaled ensembles of size up to 50 in terms of NLL and predictive set size, resulting in up to 10 times smaller inference costs. Thus, we propose JUCAL as a new go-to method for calibrating ensembles in classification.

Samsung Galaxy Unpacked 2026 is Feb. 25 The company is also said to be working on smart glasses and a smart lamp. OpenAI is reportedly hard at work developing a series of AI-powered devices, including smart glasses, a smart speaker and a smart lamp. According to reporting by, the AI company has a team of over 200 employees dedicated to the project. The first product scheduled to be released is reported to be a smart speaker that would include a camera, allowing it to better absorb information about its users and surroundings. According to a person familiar with the project, this would extend to identifying objects on a nearby table, as well as conversations being held in the vicinity of the speaker.

From left: India's prime minister, Narendra Modi, with the chief executives of OpenAI, Sam Altman, and Anthropic, Dario Amodei, at the AI Impact summit in Delhi. From left: India's prime minister, Narendra Modi, with the chief executives of OpenAI, Sam Altman, and Anthropic, Dario Amodei, at the AI Impact summit in Delhi. AI hit: India hungry to harness US tech giants' technology at Delhi summit Narendra Modi's thirst to supercharge economic growth is matched by US desire to inject AI into world's biggest democracy I ndia celebrates 80 years of independence from the UK in August 2027. At about that same moment, "early versions of true super intelligence" could emerge, Sam Altman, the co-founder of OpenAI, said this week. It's a looming coincidence that raised a charged question at the AI Impact summit in Delhi, hosted by India's prime minister, Narendra Modi: can India avoid returning to the status of a vassal state when it imports AI to raise the prospects of its 1.4 billion people? Modi's hunger to harness AI's capability is great.

India's Prime Minister Narendra Modi takes a group photo with AI company leaders at the AI Impact Summit in New Delhi on Feb. 19, 2026. India's Prime Minister Narendra Modi takes a group photo with AI company leaders at the AI Impact Summit in New Delhi on Feb. 19, 2026. The world's largest-ever AI summit took place in India this week, with hundreds of thousands of people, including world leaders and CEOs of AI companies, descending upon New Delhi for five days. It was the fourth in a series of summits that were initially designed as a place for governments to coordinate global action in the face of threats from advanced AI. But the India summit, like one in Paris before it, functioned more as a trade fair and an advertisement for the host nation's AI prowess than a venue for meaningful international diplomacy.

Indian Prime Minister Narendra Modi takes a group photo with leaders of artificial intelligence companies at the AI Impact Summit in New Delhi on Thursday. But analysts said the country was unlikely to have a "DeepSeek moment" -- the sort of boom China had last year with a high-performance, low-cost chatbot -- any time soon. Still, building custom AI tools could bring benefits to the world's most populous nation. In a time of both misinformation and too much information, quality journalism is more crucial than ever. By subscribing, you can help us get the story right. With your current subscription plan you can comment on stories.

This paper introduces a distillation framework for an ensemble of entropy-optimal Sparse Probabilistic Approximation (eSPA) models, trained exclusively on satellite-era observational and reanalysis data to predict ENSO phase up to 24 months in advance. While eSPA ensembles yield state-of-the-art forecast skill, they are harder to interpret than individual eSPA models. We show how to compress the ensemble into a compact set of "distilled" models by aggregating the structure of only those ensemble members that make correct predictions. This process yields a single, diagnostically tractable model for each forecast lead time that preserves forecast performance while also enabling diagnostics that are impractical to implement on the full ensemble. An analysis of the regime persistence of the distilled model "superclusters", as well as cross-lead clustering consistency, shows that the discretised system accurately captures the spatiotemporal dynamics of ENSO. By considering the effective dimension of the feature importance vectors, the complexity of the input space required for correct ENSO phase prediction is shown to peak when forecasts must cross the boreal spring predictability barrier. Spatial importance maps derived from the feature importance vectors are introduced to identify where predictive information resides in each field and are shown to include known physical precursors at certain lead times. Case studies of key events are also presented, showing how fields reconstructed from distilled model centroids trace the evolution from extratropical and inter-basin precursors to the mature ENSO state. Overall, the distillation framework enables a rigorous investigation of long-range ENSO predictability that complements real-time data-driven operational forecasts.

Graph neural networks (GNNs) work remarkably well in semi-supervised node regression, yet a rigorous theory explaining when and why they succeed remains lacking. To address this gap, we study an aggregate-and-readout model that encompasses several common message passing architectures: node features are first propagated over the graph then mapped to responses via a nonlinear function. For least-squares estimation over GNNs with linear graph convolutions and a deep ReLU readout, we prove a sharp non-asymptotic risk bound that separates approximation, stochastic, and optimization errors. The bound makes explicit how performance scales with the fraction of labeled nodes and graph-induced dependence. Approximation guarantees are further derived for graph-smoothing followed by smooth nonlinear readouts, yielding convergence rates that recover classical nonparametric behavior under full supervision while characterizing performance when labels are scarce. Numerical experiments validate our theory, providing a systematic framework for understanding GNN performance and limitations.

Reasoning with LLMs increasingly unfolds inside a broader verification loop. Internally, systems use cheap checks, such as self-consistency or proxy rewards, which we call weak verification. Externally, users inspect outputs and steer the model through feedback until results are trustworthy, which we call strong verification. These signals differ sharply in cost and reliability: strong verification can establish trust but is resource-intensive, while weak verification is fast and scalable but noisy and imperfect. We formalize this tension through weak--strong verification policies, which decide when to accept or reject based on weak verification and when to defer to strong verification. We introduce metrics capturing incorrect acceptance, incorrect rejection, and strong-verification frequency. Over population, we show that optimal policies admit a two-threshold structure and that calibration and sharpness govern the value of weak verifiers. Building on this, we develop an online algorithm that provably controls acceptance and rejection errors without assumptions on the query stream, the language model, or the weak verifier.

A new artificial intelligence (AI) model developed by the Chinese company behind TikTok rocked Hollywood this week - not just because of what it can do, but what it could mean for creative industries. Created by tech giant ByteDance, Seedance 2.0 can generate cinema-quality video, complete with sound effects and dialogue, from just a few written prompts. Many of the clips said to have been made using Seedance, and featuring popular characters like Spider-Man and Deadpool, went viral. What is Seedance - and why the stir? Seedance was launched to little fanfare in June 2025 but it is the second version that came eight months later that has caused a major stir.