Deep learning models have proven enormously successful at using multiple layers of representation to learn relevant features of structured data. Encoding physical symmetries into these models can improve performance on difficult tasks, and recent work has motivated the principle of parameter symmetry breaking and restoration as a unifying mechanism underlying their hierarchical learning dynamics. We evaluate the role of parameter symmetry and network expressivity in the generalisation behaviour of neural networks when learning a real-space renormalisation group (RG) transformation, using the central limit theorem (CLT) as a test case map. We consider simple multilayer perceptrons (MLPs) and graph neural networks (GNNs), and vary weight symmetries and activation functions across architectures. Our results reveal a competition between symmetry constraints and expressivity, with overly complex or overconstrained models generalising poorly. We analytically demonstrate this poor generalisation behaviour for certain constrained MLP architectures by recasting the CLT as a cumulant recursion relation and making use of an established framework to propagate cumulants through MLPs. We also empirically validate an extension of this framework from MLPs to GNNs, elucidating the internal information processing performed by these more complex models. These findings offer new insight into the learning dynamics of symmetric networks and their limitations in modelling structured physical transformations.

Modern generative models, such as neural ordinary differential equations (neural ODEs) [4], transformers [25], and diffusion models [22], have demonstrated remarkable ability to learn and generate samples from complex, high-dimensional probability distributions. These architectures have achieved broad success in scientific computing, image processing, and data science, offering scalable frameworks for data-driven modeling. However, training and sampling in such spaces remain expensive and highly sensitive to architectural and optimization choices. Despite these advances, the curse of dimensionality continues to present a fundamental challenge in many real-world applications. Fortunately, numerous problems in scientific computing exhibit intrinsic structures, such as sparsity, low-rank representations, or approximate invariances, that can be interpreted as prior information about the underlying data or operators. Leveraging such priors within generative models offers a promising avenue to improve both computational efficiency and generalization. A classical way to incorporate prior information, such as sparsity or piecewise regularity, is through Bayesian modeling, where the posterior combines a prior distribution encoding structural knowledge with a likelihood function derived from observations.

The success of deep neural networks in tasks such as image recognition, natural language processing, and reinforcement learning has come at the cost of escalating computational and memory requirements. Modern models, often comprised of billions of parameters, demand significant resources for training and inference, rendering them impractical for deployment on resource-constrained devices like mobile phones, embedded systems, or IoT devices. To address this challenge, weight quantization--reducing the precision of neural network weights--has emerged as a promising technique to lower memory footprint and accelerate inference. In particular, one-bit quantization, which restricts weights to{+1, 1}, offers extreme compression (e.g., 32 memory reduction for 32-bit floats) and enables efficient hardware implementations using bitwise operations. Various works have explored the possibility of network quantization in the recent years. In particular, for Large Language Models (LLMs), some post-training have been able to reduce the model size via fine-tuning. Examples of such approach include GPTQ Frantar et al. (2022) which can quantize a 175 billion GPT model to 4 bits and QuIP which Chee et al. (2023) compresses Llama 2 70B to 2 and 3 bits. Furthermore, quantization-aware training approaches, such as Bitnet Wang et al. (2023), Bitnet 1.58b Ma et al. (2024), have been able to achieve one-bit language models with comparable performance to the models from the same weight class. For a recent survey on efficient LLMs we refer to Xu et al. (2024).

Membership Inference Attacks (MIAs) pose a critical privacy threat by enabling adversaries to determine whether a specific sample was included in a model's training dataset. Despite extensive research on MIAs, systematic comparisons between generative and discriminative classifiers remain limited. This work addresses this gap by first providing theoretical motivation for why generative classifiers exhibit heightened susceptibility to MIAs, then validating these insights through comprehensive empirical evaluation. Our study encompasses discriminative, generative, and pseudo-generative text classifiers across varying training data volumes, evaluated on nine benchmark datasets. Employing a diverse array of MIA strategies, we consistently demonstrate that fully generative classifiers which explicitly model the joint likelihood $P(X,Y)$ are most vulnerable to membership leakage. Furthermore, we observe that the canonical inference approach commonly used in generative classifiers significantly amplifies this privacy risk. These findings reveal a fundamental utility-privacy trade-off inherent in classifier design, underscoring the critical need for caution when deploying generative classifiers in privacy-sensitive applications. Our results motivate future research directions in developing privacy-preserving generative classifiers that can maintain utility while mitigating membership inference vulnerabilities.

The recent development of foundation models for time series data has generated considerable interest in using such models across a variety of applications. Although foundation models achieve state-of-the-art predictive performance, their calibration properties remain relatively underexplored, despite the fact that calibration can be critical for many practical applications. In this paper, we investigate the calibration-related properties of five recent time series foundation models and two competitive baselines. We perform a series of systematic evaluations assessing model calibration (i.e., over- or under-confidence), effects of varying prediction heads, and calibration under long-term autoregressive forecasting. We find that time series foundation models are consistently better calibrated than baseline models and tend not to be either systematically over- or under-confident, in contrast to the overconfidence often seen in other deep learning models.

Tammy Carvey of Michigan won $100,000 in Powerball after using ChatGPT artificial intelligence to select her lottery numbers for the September drawing.

When OpenAI released its AI video-generation app, Sora, in September, it promised that "you are in control of your likeness end-to-end." The app allows users to include themselves and their friends in videos through a feature called "cameos"--the app scans a user's face and performs a liveness check, providing data to generate a video of the user and to authenticate their consent for friends to use their likeness on the app. But Reality Defender, a company specializing in identifying deepfakes, says it was able to bypass Sora's anti-impersonation safeguards within 24 hours. Platforms such as Sora give a "plausible sense of security," says Reality Defender CEO Ben Colman, despite the fact that "anybody can use completely off-the-shelf tools" to pass authentication as someone else. Reality Defender's researchers used publicly available footage of notable individuals, including CEOs and entertainers, from earnings calls and media interviews.

When you purchase through links in our articles, we may earn a small commission. Developers and creatives looking for greater control and privacy with their AI are increasingly turning to locally run models like OpenAI's new gpt-oss family of models, which are both lightweight and incredibly functional on end-user hardware. Indeed, you can have it run on consumer GPUs with just 16GB of memory. That makes it possible to use a wide range of hardware - with NVIDIA GPUs emerging as the best way to run these sorts of open-weight models. While nations and companies rush to develop their own bespoke AI solutions to a range of tasks, open source and open-weight models like OpenAI's new gpt-oss-20b are finding much more adoption.

Amazon Web Services experienced DNS resolution issues on Monday morning, taking down wide swaths of the web--and highlighting a longstanding weakness in the internet's infrastructure. A massive cloud outage stemming from Amazon Web Services's key US-EAST-1 region, its hub near the United States capitol in northern Virginia, caused widespread disruptions of websites and platforms around the world on Monday morning. Amazon's main e-commerce platform and other properties including Ring doorbells and the Alexa smart assistant suffered interruptions and outages throughout the morning, as did Meta's communication platform WhatsApp, OpenAI's ChatGPT, PayPal's Venmo payment platform, multiple web services from Epic Games, multiple British government sites, and many others. The outages stemmed from Amazon's "DynamoDB" database application programming interfaces in US-EAST-1, and AWS said in status updates that the problem was specifically related to DNS resolution issues. The "Domain Name System" is a foundational internet service that essentially acts as an automatic phonebook lookup to translate web URLs like "www.wired.com"

Plus: AWS is working to get its customers' services back online following a major outage Science Corporation--a competitor to Neuralink founded by the former president of Elon Musk's brain-interface venture--has leapfrogged its rival after acquiring a vision implant in advanced testing for a fire-sale price. The implant produces a form of "artificial vision" that lets some patients read text and do crosswords, according to a report published in today. The implant is a microelectronic chip placed under the retina. Using signals from a camera mounted on a pair of glasses, the chip emits bursts of electricity in order to bypass photoreceptor cells damaged by macular degeneration, the leading cause of vision loss in the elderly. How will flowers respond to climate change? Flowers play a key role in most landscapes, from urban to rural areas.