Images captured from a long distance suffer from dynamic image distortion due to turbulent flow of air cells with random temperatures, and thus refractive indices. This phenomenon, known as image dancing, is commonly characterized by its refractive-index structure constant $C_n^2$ as a measure of the turbulence strength. For many applications such as atmospheric forecast model, long-range/astronomy imaging, and aviation safety, optical communication technology, $C_n^2$ estimation is critical for accurately sensing the turbulent environment. Previous methods for $C_n^2$ estimation include estimation from meteorological data (temperature, relative humidity, wind shear, etc.) for single-point measurements, two-ended pathlength measurements from optical scintillometer for path-averaged $C_n^2$, and more recently estimating $C_n^2$ from passive video cameras for low cost and hardware complexity. In this paper, we present a comparative analysis of classical image gradient methods for $C_n^2$ estimation and modern deep learning-based methods leveraging convolutional neural networks. To enable this, we collect a dataset of video capture along with reference scintillometer measurements for ground truth, and we release this unique dataset to the scientific community. We observe that deep learning methods can achieve higher accuracy when trained on similar data, but suffer from generalization errors to other, unseen imagery as compared to classical methods. To overcome this trade-off, we present a novel physics-based network architecture that combines learned convolutional layers with a differentiable image gradient method that maintains high accuracy while being generalizable across image datasets.
With the increasing demand for substantial amounts of high-quality data to train large language models (LLMs), efficiently filtering large web corpora has become a critical challenge. For this purpose, KenLM, a lightweight n-gram-based language model that operates on CPUs, is widely used. However, the traditional method of training KenLM utilizes only high-quality data and, consequently, does not explicitly learn the linguistic patterns of low-quality data. To address this issue, we propose an ensemble approach that leverages two contrasting KenLMs: (i) Good KenLM, trained on high-quality data; and (ii) Bad KenLM, trained on low-quality data. Experimental results demonstrate that our approach significantly reduces noisy content while preserving high-quality content compared to the traditional KenLM training method. This indicates that our method can be a practical solution with minimal computational overhead for resource-constrained environments.
We introduce SOLAR 10.7B, a large language model (LLM) with 10.7 billion parameters, demonstrating superior performance in various natural language processing (NLP) tasks. Inspired by recent efforts to efficiently up-scale LLMs, we present a method for scaling LLMs called depth up-scaling (DUS), which encompasses depthwise scaling and continued pretraining. In contrast to other LLM up-scaling methods that use mixture-of-experts, DUS does not require complex changes to train and inference efficiently. We show experimentally that DUS is simple yet effective in scaling up high-performance LLMs from small ones. Building on the DUS model, we additionally present SOLAR 10.7B-Instruct, a variant fine-tuned for instruction-following capabilities, surpassing Mixtral-8x7B-Instruct. SOLAR 10.7B is publicly available under the Apache 2.0 license, promoting broad access and application in the LLM field.
The graph neural network (GNN) has been a powerful deep-learning tool in chemistry domain, due to its close connection with molecular graphs. Most GNN models collect and update atom and molecule features from the fed atom (and, in some cases, bond) features, which are basically based on the two-dimensional (2D) graph representation of 3D molecules. Correspondingly, the adjacency matrix, containing the information on covalent bonds, or equivalent data structures (e.g., lists) have been the main core in the feature-updating processes, such as graph convolution. However, the 2D-based models do not faithfully represent 3D molecules and their physicochemical properties, exemplified by the overlooked field effect that is a "through-space" effect, not a "through-bond" effect. The GNN model proposed herein, denoted as MolNet, is chemically intuitive, accommodating the 3D non-bond information in a molecule, with a noncovalent adjacency matrix $\bf{\bar A}$, and also bond-strength information from a weighted bond matrix $\bf{B}$. The noncovalent atoms, not directly bonded to a given atom in a molecule, are identified within 5 $\unicode{x212B}$ of cut-off range for the construction of $\bf{\bar A}$, and $\bf{B}$ has edge weights of 1, 1.5, 2, and 3 for single, aromatic, double, and triple bonds, respectively. Comparative studies show that MolNet outperforms various baseline GNN models and gives a state-of-the-art performance in the classification task of BACE dataset and regression task of ESOL dataset. This work suggests a future direction of deep-learning chemistry in the construction of deep-learning models that are chemically intuitive and comparable with the existing chemistry concepts and tools.