Search for: All records

Reconstructing a three-dimensional ocean sound speed field (SSF) from limited and noisy measurements presents an ill-posed and challenging inverse problem. Existing methods used a number of pre-specified priors (e.g., low-rank tensor and tensor neural network structures) to address this issue. However, the SSFs are often too complex to be accurately described by these pre-defined priors. While utilizing neural network-based priors trained on historical SSF data may be a viable workaround, acquiring SSF data remains a nontrivial task. This work starts with a key observation: Although natural images and SSFs admit fairly different characteristics, their denoising processes appear to share similar traits—as both remove random components from more structured signals. This observation allows us to incorporate deep denoisers trained using extensive natural images to realize zero-shot SSF reconstruction, without any extra training or network modifications. To implement this idea, an alternating direction method of multipliers (ADMM) algorithm using such a deep denoiser is proposed, which is reminiscent of the plug-and-play scheme from medical imaging. Our plug-and-play framework is tailored for SSF recovery such that the learned denoiser can be simultaneously used with other handcrafted SSF priors. Extensive numerical studies show that the new framework largely outperforms state-of-the-art baselines, especially under widely recognized challenging scenarios, e.g., when the SSF samples are taken as tensor fibers. The code is available at https://github.com/OceanSTARLab/DeepPnP. 

Eye tracking has already made its way to current commercial wearable display devices, and is becoming increasingly important for virtual and augmented reality applications. However, the existing model-based eye tracking solutions are not capable of conducting very accurate gaze angle measurements, and may not be sufficient to solve challenging display problems such as pupil steering or eyebox expansion. In this paper, we argue that accurate detection and localization of pupil in 3D space is a necessary intermediate step in model-based eye tracking. Existing methods and datasets either ignore evaluating the accuracy of 3D pupil localization or evaluate it only on synthetic data. To this end, we capture the first 3D pupilgaze-measurement dataset using a high precision setup with head stabilization and release it as the first benchmark dataset to evaluate both 3D pupil localization and gaze tracking methods. Furthermore, we utilize an advanced eye model to replace the commonly used oversimplified eye model. Leveraging the eye model, we propose a novel 3D pupil localization method with a deep learning-based corneal refraction correction. We demonstrate that our method outperforms the state-of-the-art works by reducing the 3D pupil localization error by 47.5% and the gaze estimation error by 18.7%. Our dataset and codes can be found here: link. 

The current study an empirical evaluation of the PERvasive Learning System (PERLS). PERLS is a mobile microlearning platform designed for learning anytime and anywhere, taking advantage of planned and unplanned time during a learner’s daily schedule to enhance and reinforce learning. Soldiers taking classes from the Sabalauski Air Assault School at Fort Campbell, KY were recruited. This evaluation compared the impact of PERLS on soldiers’ self-efficacy for their self-regulated learning ability. This evaluation found evidence of impact for the PERLS when implemented into classroom setting with soldiers that used PERLS indicating higher self-efficacy scores. 

Model parallelism has become a necessity for training modern large-scale deep language models. In this work, we identify a new and orthogonal dimension from existing model parallel approaches: it is possible to perform pipeline parallelism within a single training sequence for Transformer-based language models thanks to its autoregressive property. This enables a more fine-grained pipeline compared with previous work. With this key idea, we design TeraPipe, a high-performance token-level pipeline parallel algorithm for synchronous model-parallel training of Transformer-based language models. We develop a novel dynamic programming-based algorithm to calculate the optimal pipelining execution scheme given a specific model and cluster configuration. We show that TeraPipe can speed up the training by 5.0x for the largest GPT-3 model with 175 billion parameters on an AWS cluster with 48 p3.16xlarge instances compared with state-of-the-art model-parallel methods. The code for reproduction can be found at https://github.com/zhuohan123/terapipe