Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis

Contemporary models for generating images show remarkable quality and versatility. Swayed by these advantages, the research community repurposes them to generate videos. Since video content is highly redundant, we argue that naively bringing advances of image models to the video generation domain reduces motion fidelity, visual quality and impairs scalability. In this work, we build Snap Video, a video-first model that systematically addresses these challenges. To do that, we first extend the EDM framework to take into account spatially and temporally redundant pixels and naturally support video generation. Second, we show that a U-Net - a workhorse behind image generation - scales poorly when generating videos, requiring significant computational overhead. Hence, we propose a new transformer-based architecture that trains 3.31 times faster than U-Nets (and is ~4.5 faster at inference). This allows us to efficiently train a text-to-video model with billions of parameters for the first time, reach state-of-the-art results on a number of benchmarks, and generate videos with substantially higher quality, temporal consistency, and motion complexity. The user studies showed that our model was favored by a large margin over the most recent methods. See our website at https://snap-research.github.io/snapvideo/.

2D Matryoshka Sentence Embeddings

Common approaches rely on fixed-length embedding vectors from language models as sentence embeddings for downstream tasks such as semantic textual similarity (STS). Such methods are limited in their flexibility due to unknown computational constraints and budgets across various applications. Matryoshka Representation Learning (MRL) (Kusupati et al., 2022) encodes information at finer granularities, i.e., with lower embedding dimensions, to adaptively accommodate ad hoc tasks. Similar accuracy can be achieved with a smaller embedding size, leading to speedups in downstream tasks. Despite its improved efficiency, MRL still requires traversing all Transformer layers before obtaining the embedding, which remains the dominant factor in time and memory consumption. This prompts consideration of whether the fixed number of Transformer layers affects representation quality and whether using intermediate layers for sentence representation is feasible. In this paper, we introduce a novel sentence embedding model called Two-dimensional Matryoshka Sentence Embedding (2DMSE). It supports elastic settings for both embedding sizes and Transformer layers, offering greater flexibility and efficiency than MRL. We conduct extensive experiments on STS tasks and downstream applications. The experimental results demonstrate the effectiveness of our proposed model in dynamically supporting different embedding sizes and Transformer layers, allowing it to be highly adaptable to various scenarios.

Deep Hedging with Market Impact

Dynamic hedging is the practice of periodically transacting financial instruments to offset the risk caused by an investment or a liability. Dynamic hedging optimization can be framed as a sequential decision problem; thus, Reinforcement Learning (RL) models were recently proposed to tackle this task. However, existing RL works for hedging do not consider market impact caused by the finite liquidity of traded instruments. Integrating such feature can be crucial to achieve optimal performance when hedging options on stocks with limited liquidity. In this paper, we propose a novel general market impact dynamic hedging model based on Deep Reinforcement Learning (DRL) that considers several realistic features such as convex market impacts, and impact persistence through time. The optimal policy obtained from the DRL model is analysed using several option hedging simulations and compared to commonly used procedures such as delta hedging. Results show our DRL model behaves better in contexts of low liquidity by, among others: 1) learning the extent to which portfolio rebalancing actions should be dampened or delayed to avoid high costs, 2) factoring in the impact of features not considered by conventional approaches, such as previous hedging errors through the portfolio value, and the underlying asset's drift (i.e. the magnitude of its expected return).

Step-On-Feet Tuning: Scaling Self-Alignment of LLMs via Bootstrapping

Self-alignment is an effective way to reduce the cost of human annotation while ensuring promising model capability. However, most current methods complete the data collection and training steps in a single round, which may overlook the continuously improving ability of self-aligned models. This gives rise to a key query: What if we do multi-time bootstrapping self-alignment? Does this strategy enhance model performance or lead to rapid degradation? In this paper, our pioneering exploration delves into the impact of bootstrapping self-alignment on large language models. Our findings reveal that bootstrapping self-alignment markedly surpasses the single-round approach, by guaranteeing data diversity from in-context learning. To further exploit the capabilities of bootstrapping, we investigate and adjust the training order of data, which yields improved performance of the model. Drawing on these findings, we propose Step-On-Feet Tuning (SOFT) which leverages model's continuously enhanced few-shot ability to boost zero or one-shot performance. Based on easy-to-hard training recipe, we propose SOFT+ which further boost self-alignment's performance. Our experiments demonstrate the efficiency of SOFT (SOFT+) across various classification and generation tasks, highlighting the potential of bootstrapping self-alignment on continually enhancing model alignment performance.

Distributed Radiance Fields for Edge Video Compression and Metaverse Integration in Autonomous Driving

The metaverse is a virtual space that combines physical and digital elements, creating immersive and connected digital worlds. For autonomous mobility, it enables new possibilities with edge computing and digital twins (DTs) that offer virtual prototyping, prediction, and more. DTs can be created with 3D scene reconstruction methods that capture the real world's geometry, appearance, and dynamics. However, sending data for real-time DT updates in the metaverse, such as camera images and videos from connected autonomous vehicles (CAVs) to edge servers, can increase network congestion, costs, and latency, affecting metaverse services. Herein, a new method is proposed based on distributed radiance fields (RFs), multi-access edge computing (MEC) network for video compression and metaverse DT updates. RF-based encoder and decoder are used to create and restore representations of camera images. The method is evaluated on a dataset of camera images from the CARLA simulator. Data savings of up to 80% were achieved for H.264 I-frame - P-frame pairs by using RFs instead of I-frames, while maintaining high peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) qualitative metrics for the reconstructed images. Possible uses and challenges for the metaverse and autonomous mobility are also discussed.

HyperFast: Instant Classification for Tabular Data

Training deep learning models and performing hyperparameter tuning can be computationally demanding and time-consuming. Meanwhile, traditional machine learning methods like gradient-boosting algorithms remain the preferred choice for most tabular data applications, while neural network alternatives require extensive hyperparameter tuning or work only in toy datasets under limited settings. In this paper, we introduce HyperFast, a meta-trained hypernetwork designed for instant classification of tabular data in a single forward pass. HyperFast generates a task-specific neural network tailored to an unseen dataset that can be directly used for classification inference, removing the need for training a model. We report extensive experiments with OpenML and genomic data, comparing HyperFast to competing tabular data neural networks, traditional ML methods, AutoML systems, and boosting machines. HyperFast shows highly competitive results, while being significantly faster. Additionally, our approach demonstrates robust adaptability across a variety of classification tasks with little to no fine-tuning, positioning HyperFast as a strong solution for numerous applications and rapid model deployment. HyperFast introduces a promising paradigm for fast classification, with the potential to substantially decrease the computational burden of deep learning. Our code, which offers a scikit-learn-like interface, along with the trained HyperFast model, can be found at https://github.com/AI-sandbox/HyperFast.

A Language Model's Guide Through Latent Space

Concept guidance has emerged as a cheap and simple way to control the behavior of language models by probing their hidden representations for concept vectors and using them to perturb activations at inference time. While the focus of previous work has largely been on truthfulness, in this paper we extend this framework to a richer set of concepts such as appropriateness, humor, creativity and quality, and explore to what degree current detection and guidance strategies work in these challenging settings. To facilitate evaluation, we develop a novel metric for concept guidance that takes into account both the success of concept elicitation as well as the potential degradation in fluency of the guided model. Our extensive experiments reveal that while some concepts such as truthfulness more easily allow for guidance with current techniques, novel concepts such as appropriateness or humor either remain difficult to elicit, need extensive tuning to work, or even experience confusion. Moreover, we find that probes with optimal detection accuracies do not necessarily make for the optimal guides, contradicting previous observations for truthfulness. Our work warrants a deeper investigation into the interplay between detectability, guidability, and the nature of the concept, and we hope that our rich experimental test-bed for guidance research inspires stronger follow-up approaches.

A Microstructure-based Graph Neural Network for Accelerating Multiscale Simulations

Simulating the mechanical response of advanced materials can be done more accurately using concurrent multiscale models than with single-scale simulations. However, the computational costs stand in the way of the practical application of this approach. The costs originate from microscale Finite Element (FE) models that must be solved at every macroscopic integration point. A plethora of surrogate modeling strategies attempt to alleviate this cost by learning to predict macroscopic stresses from macroscopic strains, completely replacing the microscale models. In this work, we introduce an alternative surrogate modeling strategy that allows for keeping the multiscale nature of the problem, allowing it to be used interchangeably with an FE solver for any time step. Our surrogate provides all microscopic quantities, which are then homogenized to obtain macroscopic quantities of interest. We achieve this for an elasto-plastic material by predicting full-field microscopic strains using a graph neural network (GNN) while retaining the microscopic constitutive material model to obtain the stresses. This hybrid data-physics graph-based approach avoids the high dimensionality originating from predicting full-field responses while allowing non-locality to arise. By training the GNN on a variety of meshes, it learns to generalize to unseen meshes, allowing a single model to be used for a range of microstructures. The embedded microscopic constitutive model in the GNN implicitly tracks history-dependent variables and leads to improved accuracy. We demonstrate for several challenging scenarios that the surrogate can predict complex macroscopic stress-strain paths. As the computation time of our method scales favorably with the number of elements in the microstructure compared to the FE method, our method can significantly accelerate FE2 simulations.

DeepCode AI Fix: Fixing Security Vulnerabilities with Large Language Models

The automated program repair field has attracted substantial interest over the years, but despite significant research efforts, creating a system that works well for complex semantic bugs such as security vulnerabilities has proven difficult. A promising direction to solve this challenge is by leveraging large language models (LLMs), which are increasingly used to solve various programming tasks. In this paper, we investigate the effectiveness of LLMs for solving code-repair task. We show that the task is difficult as it requires the model to learn long-range code relationships, a task that inherently relies on extensive amounts of training data. At the same time, creating a large, clean dataset for complex program bugs and their corresponding fixes is non-trivial. We propose a technique to address these challenges with a new approach for querying and fine-tuning LLMs. The idea is to use program analysis to limit the LLM's attention mechanism on the portions of code needed to perform the fix, drastically reducing the amount of required training data. Concretely, for training and inference, rather than feeding the entire program to the LLM, we reduce its code to a much shorter snippet that contains the reported defect together with the necessary context - and use that instead. Our evaluation shows that this code reduction approach substantially improves available models such as GPT-4 using few-shot learning, as well as fine-tuning models. To train and evaluate our system, we created a comprehensive code fixing dataset by extensively labeling 156 bug patterns (including 40 security rules), requiring complex interprocedural dataflow to discover. Our best system with Mixtral-8x7B can remove more than 80% of the reported defects while exactly matching the human fix in between 10 and 50% of cases, outperforming baselines based on GPT-3.5 and GPT-4, or based on window-based models like TFix.

A Bound on the Maximal Marginal Degrees of Freedom

Common kernel ridge regression is expensive in memory allocation and computation time. This paper addresses low rank approximations and surrogates for kernel ridge regression, which bridge these difficulties. The fundamental contribution of the paper is a lower bound on the rank of the low dimensional approximation, which is required such that the prediction power remains reliable. The bound relates the effective dimension with the largest statistical leverage score. We characterize the effective dimension and its growth behavior with respect to the regularization parameter by involving the regularity of the kernel. This growth is demonstrated to be asymptotically logarithmic for suitably chosen kernels, justifying low-rank approximations as the Nystr\"om method.