scholarly journals Efficient knowledge distillation for liver CT segmentation using growing assistant network

2021 ◽  
Author(s):  
Pengcheng Xu ◽  
Kyungsang Kim ◽  
Jeongwan Koh ◽  
Dufan Wu ◽  
Yu Rim Lee ◽  
...  
Keyword(s):  
Knowledge Transfer ◽  
Real Time ◽  
Liver Surgery ◽  
Computational Cost ◽  
Lesion Detection ◽  
Dice Similarity Coefficient ◽  
Student Model ◽  
Liver Ct ◽  
Knowledge Distillation ◽  
Ct Segmentation

Abstract Segmentation has been widely used in diagnosis, lesion detection, and surgery planning. Although deep learning (DL)-based segmentation methods currently outperform traditional methods, most DL-based segmentation models are computationally expensive and memory inefficient, which are not suitable for the intervention of liver surgery. To address this issue, a simple solution is to make a segmentation model very small for the fast inference time, however, there is a trade-off between the model size and performance. In this paper, we propose a DL-based real- time 3-D liver CT segmentation method, where knowledge distillation (KD) method, referred to as knowledge transfer from teacher to student models, is incorporated to compress the model while preserving the performance. Because it is known that the knowledge transfer is inefficient when the disparity of teacher and student model sizes is large, we propose a growing teacher assistant network (GTAN) to gradually learn the knowledge without extra computational cost, which can efficiently transfer knowledges even with the large gap of teacher and student model sizes. In our results, dice similarity coefficient of the student model with KD improved 1.2% (85.9% to 87.1%) compared to the student model without KD, which is a similar performance of the teacher model using only 8% (100k) parameters. Furthermore, with a student model of 2% (30k) parameters, the proposed model using the GTAN improved the dice coefficient about 2% compared to the student model without KD, with the inference time of 13ms per case. Therefore, the proposed method has a great potential for intervention in liver surgery, which also can be utilized in many real-time applications.

scholarly journals Online Knowledge Distillation with Diverse Peers

2020 ◽  
Vol 34 (04) ◽  
pp. 3430-3437
Author(s):  
Defang Chen ◽  
Jian-Ping Mei ◽  
Can Wang ◽  
Yan Feng ◽  
Chun Chen
Keyword(s):  
Knowledge Transfer ◽  
State Of The Art ◽  
High Capacity ◽  
Group Leader ◽  
Student Model ◽  
Aggregation Functions ◽  
Knowledge Distillation ◽  
Group Members ◽  
Student Models ◽  
Soft Targets

Distillation is an effective knowledge-transfer technique that uses predicted distributions of a powerful teacher model as soft targets to train a less-parameterized student model. A pre-trained high capacity teacher, however, is not always available. Recently proposed online variants use the aggregated intermediate predictions of multiple student models as targets to train each student model. Although group-derived targets give a good recipe for teacher-free distillation, group members are homogenized quickly with simple aggregation functions, leading to early saturated solutions. In this work, we propose Online Knowledge Distillation with Diverse peers (OKDDip), which performs two-level distillation during training with multiple auxiliary peers and one group leader. In the first-level distillation, each auxiliary peer holds an individual set of aggregation weights generated with an attention-based mechanism to derive its own targets from predictions of other auxiliary peers. Learning from distinct target distributions helps to boost peer diversity for effectiveness of group-based distillation. The second-level distillation is performed to transfer the knowledge in the ensemble of auxiliary peers further to the group leader, i.e., the model used for inference. Experimental results show that the proposed framework consistently gives better performance than state-of-the-art approaches without sacrificing training or inference complexity, demonstrating the effectiveness of the proposed two-level distillation framework.

Sky Segmentation for Enhanced Depth Reconstruction and Bokeh Rendering with Efficient Architectures

2020 ◽  
Vol 2020 (14) ◽  
pp. 378-1-378-7
Author(s):  
Tyler Nuanes ◽  
Matt Elsey ◽  
Radek Grzeszczuk ◽  
John Paul Shen
Keyword(s):  
Real Time ◽  
Mobile Device ◽  
Computational Cost ◽  
False Positives ◽  
Compact Model ◽  
High Quality ◽  
False Negatives ◽  
Trade Off ◽  
Depth Reconstruction ◽  
Binary Classifiers

We present a high-quality sky segmentation model for depth refinement and investigate residual architecture performance to inform optimally shrinking the network. We describe a model that runs in near real-time on mobile device, present a new, highquality dataset, and detail a unique weighing to trade off false positives and false negatives in binary classifiers. We show how the optimizations improve bokeh rendering by correcting stereo depth misprediction in sky regions. We detail techniques used to preserve edges, reject false positives, and ensure generalization to the diversity of sky scenes. Finally, we present a compact model and compare performance of four popular residual architectures (ShuffleNet, MobileNetV2, Resnet-101, and Resnet-34-like) at constant computational cost.

scholarly journals Real-Time Path Planning Based on Harmonic Functions under a Proper Generalized Decomposition-Based Framework

Sensors ◽  
2021 ◽  
Vol 21 (12) ◽  
pp. 3943
Author(s):  
Nicolas Montés ◽  
Francisco Chinesta ◽  
Marta C. Mora ◽  
Antonio Falcó ◽  
Lucia Hilario ◽  
...  
Keyword(s):  
Path Planning ◽  
Real Time ◽  
Harmonic Functions ◽  
Autonomous Navigation ◽  
Numerical Technique ◽  
Computational Cost ◽  
Main Idea ◽  
Main Property ◽  
Proper Generalized Decomposition ◽  
Computational Vademecum

This paper presents a real-time global path planning method for mobile robots using harmonic functions, such as the Poisson equation, based on the Proper Generalized Decomposition (PGD) of these functions. The main property of the proposed technique is that the computational cost is negligible in real-time, even if the robot is disturbed or the goal is changed. The main idea of the method is the off-line generation, for a given environment, of the whole set of paths from any start and goal configurations of a mobile robot, namely the computational vademecum, derived from a harmonic potential field in order to use it on-line for decision-making purposes. Up until now, the resolution of the Laplace or Poisson equations has been based on traditional numerical techniques unfeasible for real-time calculation. This drawback has prevented the extensive use of harmonic functions in autonomous navigation, despite their powerful properties. The numerical technique that reverses this situation is the Proper Generalized Decomposition. To demonstrate and validate the properties of the PGD-vademecum in a potential-guided path planning framework, both real and simulated implementations have been developed. Simulated scenarios, such as an L-Shaped corridor and a benchmark bug trap, are used, and a real navigation of a LEGO®MINDSTORMS robot running in static environments with variable start and goal configurations is shown. This device has been selected due to its computational and memory-restricted capabilities, and it is a good example of how its properties could help the development of social robots.

scholarly journals An Iterative Multi-Source Mutual Knowledge Transfer Framework for Machine Reading Comprehension

2020 ◽  
Author(s):  
Xin Liu ◽  
Kai Liu ◽  
Xiang Li ◽  
Jinsong Su ◽  
Yubin Ge ◽  
...  
Keyword(s):  
Reading Comprehension ◽  
Knowledge Transfer ◽  
Training Data ◽  
Target Domain ◽  
Domain Specific ◽  
Mutual Knowledge ◽  
Benchmark Datasets ◽  
Knowledge Distillation ◽  
The Many ◽  
Machine Reading

The lack of sufficient training data in many domains, poses a major challenge to the construction of domain-specific machine reading comprehension (MRC) models with satisfying performance. In this paper, we propose a novel iterative multi-source mutual knowledge transfer framework for MRC. As an extension of the conventional knowledge transfer with one-to-one correspondence, our framework focuses on the many-to-many mutual transfer, which involves synchronous executions of multiple many-to-one transfers in an iterative manner.Specifically, to update a target-domain MRC model, we first consider other domain-specific MRC models as individual teachers, and employ knowledge distillation to train a multi-domain MRC model, which is differentially required to fit the training data and match the outputs of these individual models according to their domain-level similarities to the target domain. After being initialized by the multi-domain MRC model, the target-domain MRC model is fine-tuned to match both its training data and the output of its previous best model simultaneously via knowledge distillation. Compared with previous approaches, our framework can continuously enhance all domain-specific MRC models by enabling each model to iteratively and differentially absorb the domain-shared knowledge from others. Experimental results and in-depth analyses on several benchmark datasets demonstrate the effectiveness of our framework.

Development of a Fast Fluid Dynamics Model Based on PISO Algorithm for Simulating Indoor Airflow

2021 ◽  
Author(s):  
Sibo Li ◽  
Hongtao Qiao
Keyword(s):  
Fluid Dynamics ◽  
Real Time ◽  
Computational Cost ◽  
Flow Simulation ◽  
Building Design ◽  
Energy Performance ◽  
Computational Effort ◽  
Computational Method ◽  
Dimensional Case ◽  
Time Flow

Abstract Real-time or faster-than-real-time flow simulation is crucial for studying airflow and heat transfer in buildings, such as building design, building emergency management and building energy performance evaluation. Computational Fluid Dynamics (CFD) with Pressure Implicit with Splitting of Operator (PISO) or Semi-Implicit Method for Pressure Linked Equations (SIMPLE) algorithm is accurate but requires great computational resources. Fast Fluid Dynamics (FFD) can reduce the computational effort but generally lack prediction accuracy due to simplification. This study developed a fast computational method based on FFD in combination with the PISO algorithm. Boussinesq approximation is adopted for simulating buoyancy effect. The proposed solver is tested in a two-dimensional case and a three-dimensional case with experimental data. The predicted results have good agreement with the experimental results. In the two test cases, the proposed solver generates lower Root Mean Square Error (RMSE) compared to the FFD and at the same time, the proposed method reduces computational cost by a factor of 10 and 13 in the two cases compared to CFD.

Robust cross-lingual knowledge base question answering via knowledge distillation

2021 ◽  
Vol ahead-of-print (ahead-of-print) ◽  
Author(s):  
Shaofei Wang ◽  
Depeng Dang
Keyword(s):  
Knowledge Base ◽  
Design Methodology ◽  
Question Answering ◽  
Ground Truth ◽  
Student Model ◽  
Content Type ◽  
Final Performance ◽  
Knowledge Distillation ◽  
The Cross ◽  
Cross Lingual

PurposePrevious knowledge base question answering (KBQA) models only consider the monolingual scenario and cannot be directly extended to the cross-lingual scenario, in which the language of questions and that of knowledge base (KB) are different. Although a machine translation (MT) model can bridge the gap through translating questions to the language of KB, the noises of translated questions could accumulate and further sharply impair the final performance. Therefore, the authors propose a method to improve the robustness of KBQA models in the cross-lingual scenario.Design/methodology/approachThe authors propose a knowledge distillation-based robustness enhancement (KDRE) method. Specifically, first a monolingual model (teacher) is trained by ground truth (GT) data. Then to imitate the practical noises, a noise-generating model is designed to inject two types of noise into questions: general noise and translation-aware noise. Finally, the noisy questions are input into the student model. Meanwhile, the student model is jointly trained by GT data and distilled data, which are derived from the teacher when feeding GT questions.FindingsThe experimental results demonstrate that KDRE can improve the performance of models in the cross-lingual scenario. The performance of each module in KBQA model is improved by KDRE. The knowledge distillation (KD) and noise-generating model in the method can complementarily boost the robustness of models.Originality/valueThe authors first extend KBQA models from monolingual to cross-lingual scenario. Also, the authors first implement KD for KBQA to develop robust cross-lingual models.

The clinical potential of ultra-high-speed echo-planar imaging

1990 ◽  
Vol 333 (1632) ◽  
pp. 507-514 ◽  
Keyword(s):  
Real Time ◽  
High Speed ◽  
Echo Planar Imaging ◽  
Lesion Detection ◽  
Whole Body ◽  
Planar Imaging ◽  
Cardiac Events ◽  
Ultra High Speed ◽  
Movement Artefact ◽  
Echo Planar

Ultra-high-speed echo-planar imaging (EP1) allows acquisition of a complete twodimensional image in 64 to 128 ms devoid of movement artefact and without sacrifice of contrast due to relaxation time effects. In conventional whole-body MRI, however, obtrusive movement artefact and extended imaging time, resulting from the need to apply multiple sequences to facilitate lesion detection and pathological characterization, remain limitations. Reduced total examination time increases patient tolerance and throughput • furthermore optimization of contrast to achieve maximal conspicuity of particular features in liver or brain pathology is achieved simply and interactively by real time adjustment of the imaging parameters. The method provides the opportunity to study in real time dynamic events such as flow phenomena in the vascular and cerebrospinal fluid compartments of the brain as well as the kinetics of administered contrast agents, EPI is the only means of capturing the irregular motion of aperiodic cardiac events and bowel peristalsis.

scholarly journals Towards Real-Time Reinforcement Learning Control of a Wave Energy Converter

2020 ◽  
Vol 8 (11) ◽  
pp. 845
Author(s):  
Enrico Anderlini ◽  
Salman Husain ◽  
Gordon G. Parker ◽  
Mohammad Abusara ◽  
Giles Thomas
Keyword(s):  
Reinforcement Learning ◽  
Model Predictive Control ◽  
System Dynamics ◽  
Real Time ◽  
Predictive Control ◽  
Wave Energy ◽  
Economies Of Scale ◽  
Computational Cost ◽  
Effective Control ◽  
Deployment Time

The levellised cost of energy of wave energy converters (WECs) is not competitive with fossil fuel-powered stations yet. To improve the feasibility of wave energy, it is necessary to develop effective control strategies that maximise energy absorption in mild sea states, whilst limiting motions in high waves. Due to their model-based nature, state-of-the-art control schemes struggle to deal with model uncertainties, adapt to changes in the system dynamics with time, and provide real-time centralised control for large arrays of WECs. Here, an alternative solution is introduced to address these challenges, applying deep reinforcement learning (DRL) to the control of WECs for the first time. A DRL agent is initialised from data collected in multiple sea states under linear model predictive control in a linear simulation environment. The agent outperforms model predictive control for high wave heights and periods, but suffers close to the resonant period of the WEC. The computational cost at deployment time of DRL is also much lower by diverting the computational effort from deployment time to training. This provides confidence in the application of DRL to large arrays of WECs, enabling economies of scale. Additionally, model-free reinforcement learning can autonomously adapt to changes in the system dynamics, enabling fault-tolerant control.

scholarly journals A Deep Learning Method for Near-Real-Time Cloud and Cloud Shadow Segmentation from Gaofen-1 Images

2020 ◽  
Vol 2020 ◽  
pp. 1-13
Author(s):  
Mehdi Khoshboresh-Masouleh ◽  
Reza Shah-Hosseini
Keyword(s):  
Deep Learning ◽  
Real Time ◽  
Computational Cost ◽  
High Definition ◽  
Satellite Mission ◽  
Real Time Processing ◽  
National Space ◽  
Spatial Features ◽  
Efficiency And Effectiveness ◽  
Earth Observation Satellite

In this study, an essential application of remote sensing using deep learning functionality is presented. Gaofen-1 satellite mission, developed by the China National Space Administration (CNSA) for the civilian high-definition Earth observation satellite program, provides near-real-time observations for geographical mapping, environment surveying, and climate change monitoring. Cloud and cloud shadow segmentation are a crucial element to enable automatic near-real-time processing of Gaofen-1 images, and therefore, their performances must be accurately validated. In this paper, a robust multiscale segmentation method based on deep learning is proposed to improve the efficiency and effectiveness of cloud and cloud shadow segmentation from Gaofen-1 images. The proposed method first implements feature map based on the spectral-spatial features from residual convolutional layers and the cloud/cloud shadow footprints extraction based on a novel loss function to generate the final footprints. The experimental results using Gaofen-1 images demonstrate the more reasonable accuracy and efficient computational cost achievement of the proposed method compared to the cloud and cloud shadow segmentation performance of two existing state-of-the-art methods.

scholarly journals HyP-DESPOT: A hybrid parallel algorithm for online planning under uncertainty

2020 ◽  
pp. 027836492093707
Author(s):  
Panpan Cai ◽  
Yuanfu Luo ◽  
David Hsu ◽  
Wee Sun Lee
Keyword(s):  
Real Time ◽  
Computational Cost ◽  
Search Tree ◽  
Planning Under Uncertainty ◽  
State Action ◽  
Online Planning ◽  
Planning Algorithm ◽  
One Step ◽  
Robotic Tasks ◽  
High Computational Cost

Robust planning under uncertainty is critical for robots in uncertain, dynamic environments, but incurs high computational cost. State-of-the-art online search algorithms, such as DESPOT, have vastly improved the computational efficiency of planning under uncertainty and made it a valuable tool for robotics in practice. This work takes one step further by leveraging both CPU and GPU parallelization in order to achieve real-time online planning performance for complex tasks with large state, action, and observation spaces. Specifically, Hybrid Parallel DESPOT (HyP-DESPOT) is a massively parallel online planning algorithm that integrates CPU and GPU parallelism in a multi-level scheme. It performs parallel DESPOT tree search by simultaneously traversing multiple independent paths using multi-core CPUs; it performs parallel Monte Carlo simulations at the leaf nodes of the search tree using GPUs. HyP-DESPOT provably converges in finite time under moderate conditions and guarantees near-optimality of the solution. Experimental results show that HyP-DESPOT speeds up online planning by up to a factor of several hundred in several challenging robotic tasks in simulation, compared with the original DESPOT algorithm. It also exhibits real-time performance on a robot vehicle navigating among many pedestrians.

Sign in / Sign up

Export Citation Format

Share Document