We present a smoothly broken power law functional form (that we refer to as a Broken Neural Scaling Law (BNSL)) that accurately models & extrapolates the scaling behaviors of deep neural networks (i.e. how the evaluation metric of interest varies as amount of compute used for training (or inference), number of model parameters, training dataset size, model input size, number of training steps, or upstream performance varies) for various architectures & for each of various tasks within a large & diverse set of upstream & downstream tasks, in zero-shot, prompted, & finetuned settings. This set includes large-scale vision, language, audio, video, diffusion, generative modeling, multimodal learning, contrastive learning, AI alignment, AI capabilities, robotics, out-of-distribution (OOD) generalization, continual learning, transfer learning, uncertainty estimation / calibration, OOD detection, adversarial robustness, distillation, sparsity, retrieval, quantization, pruning, fairness, molecules, computer programming/coding, math word problems, "emergent phase transitions", arithmetic, supervised learning, unsupervised/self-supervised learning, & reinforcement learning (single agent & multi-agent). When compared to other functional forms for neural scaling, this functional form yields extrapolations of scaling behavior that are considerably more accurate on this set. Moreover, this functional form accurately models & extrapolates scaling behavior that other functional forms are incapable of expressing such as the nonmonotonic transitions present in the scaling behavior of phenomena such as double descent & the delayed, sharp inflection points present in the scaling behavior of tasks such as arithmetic. Lastly, we use this functional form to glean insights about the limit of the predictability of scaling behavior. Code is available at https://github.com/ethancaballero/broken_neural_scaling_laws

Combining off-the-shelf components with 3D-printing, the Wheelbot is a symmetric reaction wheel unicycle that can jump onto its wheels from any initial position. With non-holonomic and under-actuated dynamics, as well as two coupled unstable degrees of freedom, the Wheelbot provides a challenging platform for nonlinear and data-driven control research. This paper presents the Wheelbot's mechanical and electrical design, its estimation and control algorithms, as well as experiments demonstrating both self-erection and disturbance rejection while balancing.

The biggest Women's World Cup ever kicked off in Australia and New Zealand earlier this week with wins for both of the host nations. For the first time, the women's version of FIFA's tournament has 32 teams participating, following the format of the men's competition for the past 25 years. It comprises 64 matches across five time zones in nine cities, culminating with the final in Sydney on August 20. On Saturday, England kick off their campaign with a match against Haiti in Brisbane at 7:30pm local time (10:30am BST). MailOnline has taken a look at the innovations underpinning the player and fan experience this year, including AI-powered limb-tracking, a new video assistant referee procedure and a Web3 prediction game.

Wireless capsule endoscopy (WCE) is a painless and non-invasive diagnostic tool for gastrointestinal (GI) diseases. However, due to GI anatomical constraints and hardware manufacturing limitations, WCE vision signals may suffer from insufficient illumination, leading to a complicated screening and examination procedure. Deep learning-based low-light image enhancement (LLIE) in the medical field gradually attracts researchers. Given the exuberant development of the denoising diffusion probabilistic model (DDPM) in computer vision, we introduce a WCE LLIE framework based on the multi-scale convolutional neural network (CNN) and reverse diffusion process. The multi-scale design allows models to preserve high-resolution representation and context information from low-resolution, while the curved wavelet attention (CWA) block is proposed for high-frequency and local feature learning. Furthermore, we combine the reverse diffusion procedure to further optimize the shallow output and generate the most realistic image. The proposed method is compared with ten state-of-the-art (SOTA) LLIE methods and significantly outperforms quantitatively and qualitatively. The superior performance on GI disease segmentation further demonstrates the clinical potential of our proposed model. Our code is publicly accessible.

Advancements in generative artificial intelligence (AI) and large language models (LLMs) have fueled the development of many educational technology innovations that aim to automate the often time-consuming and laborious tasks of generating and analysing textual content (e.g., generating open-ended questions and analysing student feedback survey) (Kasneci et al., 2023; Wollny et al., 2021; Leiker et al., 2023). LLMs are generative artificial intelligence models that have been trained on an extensive amount of text data, capable of generating human-like text content based on natural language inputs. Specifically, these LLMs, such as Bidirectional Encoder Representations from Transformers (BERT) (Devlin et al., 2018) and Generative Pre-trained Transformer (GPT) (Brown et al., 2020), utilise deep learning and self-attention mechanisms (Vaswani et al., 2017) to selectively attend to the different parts of input texts, depending on the focus of the current tasks, allowing the model to learn complex patterns and relationships among textual contents, such as their semantic, contextual, and syntactic relationships (Min et al., 2021; Liu et al., 2023). As several LLMs (e.g., GPT-3 and Codex) have been pre-trained on massive amounts of data across multiple disciplines, they are capable of completing natural language processing tasks with little (few-shot learning) or no additional training (zero-shot learning) (Brown et al., 2020; Wu et al., 2023). This could lower the technological barriers to LLMs-based innovations as researchers and practitioners can develop new educational technologies by fine-tuning LLMs on specific educational tasks without starting from scratch (Caines et al., 2023; Sridhar et al., 2023). The recent release of ChatGPT, an LLMs-based generative AI chatbot that requires only natural language prompts without additional model training or fine-tuning (OpenAI, 2023), has further lowered the barrier for individuals without technological background to leverage the generative powers of LLMs. Although educational research that leverages LLMs to develop technological innovations for automating educational tasks is yet to achieve its full potential (i.e., most works have focused on improving model performances (Kurdi et al., 2020; Ramesh and Sanampudi, 2022)), a growing body of literature hints at how different stakeholders could potentially benefit from such innovations.

Identifying Amyotrophic Lateral Sclerosis (ALS) in its early stages is essential for establishing the beginning of treatment, enriching the outlook, and enhancing the overall well-being of those affected individuals. However, early diagnosis and detecting the disease's signs is not straightforward. A simpler and cheaper way arises by analyzing the patient's facial expressions through computational methods. When a patient with ALS engages in specific actions, e.g., opening their mouth, the movement of specific facial muscles differs from that observed in a healthy individual. This paper proposes Facial Point Graphs to learn information from the geometry of facial images to identify ALS automatically. The experimental outcomes in the Toronto Neuroface dataset show the proposed approach outperformed state-of-the-art results, fostering promising developments in the area.

As mobile devices become increasingly popular for video streaming, it's crucial to optimize the streaming experience for these devices. Although deep learning-based video enhancement techniques are gaining attention, most of them cannot support real-time enhancement on mobile devices. Additionally, many of these techniques are focused solely on super-resolution and cannot handle partial or complete loss or corruption of video frames, which is common on the Internet and wireless networks. To overcome these challenges, we present a novel approach in this paper. Our approach consists of (i) a novel video frame recovery scheme, (ii) a new super-resolution algorithm, and (iii) a receiver enhancement-aware video bit rate adaptation algorithm. We have implemented our approach on an iPhone 12, and it can support 30 frames per second (FPS). We have evaluated our approach in various networks such as WiFi, 3G, 4G, and 5G networks. Our evaluation shows that our approach enables real-time enhancement and results in a significant increase in video QoE (Quality of Experience) of 24\% - 82\% in our video streaming system.

Accident detection and traffic analysis is a critical component of smart city and autonomous transportation systems that can reduce accident frequency, severity and improve overall traffic management. This paper presents a comprehensive analysis of traffic accidents in different regions across the United States using data from the National Highway Traffic Safety Administration (NHTSA) Crash Report Sampling System (CRSS). To address the challenges of accident detection and traffic analysis, this paper proposes a framework that uses traffic surveillance cameras and action recognition systems to detect and respond to traffic accidents spontaneously. Integrating the proposed framework with emergency services will harness the power of traffic cameras and machine learning algorithms to create an efficient solution for responding to traffic accidents and reducing human errors. Advanced intelligence technologies, such as the proposed accident detection systems in smart cities, will improve traffic management and traffic accident severity. Overall, this study provides valuable insights into traffic accidents in the US and presents a practical solution to enhance the safety and efficiency of transportation systems.

Knowledge graphs can be used in a wide range of areas which require data semantics such as question-answering systems, semantic search systems, and knowledge based systems. A knowledge graph [1, 2, 3] can be constructed using data sources from an open collaboration platform such as wikipedia or wikidata because an enormous amount of information can be gathered in the open collaboration platform. However, the constructed knowledge graph is still incomplete because there can exist a much larger number of potential relations (i.e., N N R, N: the number of entities, R: the number of relation types) compared with the number of relations in the existing knowledge graph and data sources from the open platform intrinsically cannot have all the information to connect the relations. Therefore, we need to add a lot of missing relations (or links) to the knowledge graph. It is called knowledge graph completion. Knowledge graph embedding is one of the most commonly used techniques for knowledge graph completion. Much work [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] has been studied in the literature to improve the accuracy of knowledge graph completion. However, most of the knowledge graph embedding studies do not tackle the running time of the knowledge graph completion. To find a meaningful link (i.e., to add a new relation to the knowledge graph), we should compute the score of each triplet (head, relation, tail) and the number of triplets to be computed is very huge (i.e., N N R, N: is the number of nodes, R is the number of relation types).

Vision-and-language navigation (VLN) requires an embodied agent to navigate in realistic 3D environments using natural language instructions. Existing VLN methods suffer from training on small-scale environments or unreasonable path-instruction datasets, limiting the generalization to unseen environments. There are massive house tour videos on YouTube, providing abundant real navigation experiences and layout information. However, these videos have not been explored for VLN before. In this paper, we propose to learn an agent from these videos by creating a large-scale dataset which comprises reasonable path-instruction pairs from house tour videos and pre-training the agent on it. To achieve this, we have to tackle the challenges of automatically constructing path-instruction pairs and exploiting real layout knowledge from raw and unlabeled videos. To address these, we first leverage an entropy-based method to construct the nodes of a path trajectory. Then, we propose an action-aware generator for generating instructions from unlabeled trajectories. Last, we devise a trajectory judgment pretext task to encourage the agent to mine the layout knowledge. Experimental results show that our method achieves state-of-the-art performance on two popular benchmarks (R2R and REVERIE). Code is available at https://github.com/JeremyLinky/YouTube-VLN