Fast and accurate simulation of dynamical systems is a fundamental challenge across scientific and engineering domains. Traditional numerical integrators often face a trade-off between accuracy and computational efficiency, while existing neural network-based approaches typically require training a separate model for each case. To overcome these limitations, we introduce a novel multi-modal foundation model for large-scale simulations of differential equations: FMint-SDE (Foundation Model based on Initialization for stochastic differential equations). Based on a decoder-only transformer with in-context learning, FMint-SDE leverages numerical and textual modalities to learn a universal error-correction scheme. It is trained using prompted sequences of coarse solutions generated by conventional solvers, enabling broad generalization across diverse systems. We evaluate our models on a suite of challenging SDE benchmarks spanning applications in molecular dynamics, mechanical systems, finance, and biology. Experimental results show that our approach achieves a superior accuracy-efficiency tradeoff compared to classical solvers, underscoring the potential of FMint-SDE as a general-purpose simulation tool for dynamical systems.

As SpaceX's Starship vehicle gathered all of the attention this week, the company's workhorse Falcon 9 rocket continued to hit some impressive milestones. Both occurred during relatively anonymous launches of the company's Starlink satellites but are nonetheless notable because they underscore the value of first-stage reuse, which SpaceX has pioneered over the past decade. The first milestone occurred on Wednesday morning with the launch of the Starlink 10-56 mission from Cape Canaveral, Florida. The first stage that launched these satellites, Booster 1096, was making its second launch and successfully landed on the Just Read the Instructions drone ship. Strikingly, this was the 400th time SpaceX has executed a drone ship landing.

Datasets in engineering domains are often small, sparsely labeled, and contain numerical as well as categorical conditions. Additionally. computational resources are typically limited in practical applications which hinders the adoption of generative models for engineering tasks. We introduce a novel masked-conditioning approach, that enables generative models to work with sparse, mixed-type data. We mask conditions during training to simulate sparse conditions at inference time. For this purpose, we explore the use of various sparsity schedules that show different strengths and weaknesses. In addition, we introduce a flexible embedding that deals with categorical as well as numerical conditions. We integrate our method into an efficient variational autoencoder as well as a latent diffusion model and demonstrate the applicability of our approach on two engineering-related datasets of 2D point clouds and images. Finally, we show that small models trained on limited data can be coupled with large pretrained foundation models to improve generation quality while retaining the controllability induced by our conditioning scheme.

Large Language Models have emerged as prime candidates to tackle misinformation mitigation. However, existing approaches struggle with hallucinations and overconfident predictions. We propose an uncertainty quantification framework that leverages both direct confidence elicitation and sampled-based consistency methods to provide better calibration for NLP misinformation mitigation solutions. We first investigate the calibration of sample-based consistency methods that exploit distinct features of consistency across sample sizes and stochastic levels. Next, we evaluate the performance and distributional shift of a robust numeric verbalization prompt across single vs. two-step confidence elicitation procedure. We also compare the performance of the same prompt with different versions of GPT and different numerical scales. Finally, we combine the sample-based consistency and verbalized methods to propose a hybrid framework that yields a better uncertainty estimation for GPT models. Overall, our work proposes novel uncertainty quantification methods that will improve the reliability of Large Language Models in misinformation mitigation applications.

Such Swapped errors are also common vehicles, object detection remains fairly in many classification datasets (Northcutt et al., 2021a), brittle in part due to annotation errors that plague but the increased complexity of object detection annotation most real-world training datasets. We propose introduces potential for more varied types of label errors ObjectLab, a straightforward algorithm to detect than encountered in classification. We propose an algorithm, diverse errors in object detection labels, including: ObjectLab, that utilizes any trained object detection model overlooked bounding boxes, badly located boxes, to estimate the incorrect labels in such a dataset, regardless and incorrect class label assignments. Object-which of these 3 types of mistake the data annotators made. Lab utilizes any trained object detection model to score the label quality of each image, such that Training and evaluating models with incorrect bounding box mislabeled images can be automatically prioritized annotations is clearly worrisome.

The labor-intensive annotation process of semantic segmentation datasets is often prone to errors, since humans struggle to label every pixel correctly. We study algorithms to automatically detect such annotation errors, in particular methods to score label quality, such that the images with the lowest scores are least likely to be correctly labeled. This helps prioritize what data to review in order to ensure a high-quality training/evaluation dataset, which is critical in sensitive applications such as medical imaging and autonomous vehicles. Widely applicable, our label quality scores rely on probabilistic predictions from a trained segmentation model -- any model architecture and training procedure can be utilized. Here we study 7 different label quality scoring methods used in conjunction with a DeepLabV3+ or a FPN segmentation model to detect annotation errors in a version of the SYNTHIA dataset. Precision-recall evaluations reveal a score -- the soft-minimum of the model-estimated likelihoods of each pixel's annotated class -- that is particularly effective to identify images that are mislabeled, across multiple types of annotation error.

Noise plagues many numerical datasets, where the recorded values in the data may fail to match the true underlying values due to reasons including: erroneous sensors, data entry/processing mistakes, or imperfect human estimates. Here we consider estimating which data values are incorrect along a numerical column. We present a model-agnostic approach that can utilize any regressor (i.e. statistical or machine learning model) which was fit to predict values in this column based on the other variables in the dataset. By accounting for various uncertainties, our approach distinguishes between genuine anomalies and natural data fluctuations, conditioned on the available information in the dataset. We establish theoretical guarantees for our method and show that other approaches like conformal inference struggle to detect errors. We also contribute a new error detection benchmark involving 5 regression datasets with real-world numerical errors (for which the true values are also known). In this benchmark and additional simulation studies, our method identifies incorrect values with better precision/recall than other approaches.

Generative artificial intelligence company Stability AI Ltd. today released an updated version of its popular open-source photorealistic image generation model. London-based Stability AI is the developer of Stable Diffusion, an AI model that can automatically generate realistic-looking images based on text prompts. It's not the only AI model with that capability, but what differentiates it is that it's available under an open-source license and can run on relatively simple hardware. These two features have helped it quickly amass a large user base. The latest model is called Stable Diffusion XL, and it's the latest addition to the Stable Diffusion suite.

This week: TikTok challenges Google and Microsoft with search ads, GPT-4 is on the way, and social media engagement rates are dropping. Here's what happened this week in digital marketing. OpenAI hasn't been in the news enough lately so it's time for a fresh update. The next version of GPT, unimaginatively called GPT-4, will go live soon. In fact, it might already be live by the time you read this. As far as the updates that make it more worthwhile than GPT-3, it's got multimodal functionality. That means it supports text, speech, images, and even video. GPT-4 also works across multiple languages. If you've noticed that your social media engagement rates are on the decline, you're not alone.

Nvidia Corp. announced at its virtual GTC 2022 event today that the first products and services based on its next-generation graphics processing unit, the Nvidia H100 Tensor Core GPU, will roll out next month. The Nvidia H100 Tensor Core is the most powerful GPU the company has ever made. Now in full production, it's based on the new Hopper architecture and is packed with more than 80 billion transistors. It also has new features such as a Transformer Engine and more scalable NVLink interconnect, enabling it to power larger artificial intelligence models, recommendation systems and other kinds of workloads. When the chip was first announced in April, Nvidia said it is so powerful that just 20 of them could theoretically be used to sustain the world's entire internet traffic.