Recent Efficiency Gains in Deep Learning: Performance, Power, and Sustainability

2021 ◽  
Author(s):  
Miro Hodak ◽  
Ajay Dholakia
Keyword(s):  
Deep Learning ◽  
Learning Performance ◽  
Efficiency Gains

Machine Learning as a Silent Observer of Advanced Geoscience Interpretation

2021 ◽  
Author(s):  
Muhammad Sajid
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Transfer Learning ◽  
Learning Performance ◽  
Learning Approach ◽  
Seismic Fault ◽  
Facies Classification ◽  
The World ◽  
Learning Machine ◽  
Fault Interpretation

Abstract Machine learning is proving its successes in all fields of life including medical, automotive, planning, engineering, etc. In the world of geoscience, ML showed impressive results in seismic fault interpretation, advance seismic attributes analysis, facies classification, and geobodies extraction such as channels, carbonates, and salt, etc. One of the challenges faced in geoscience is the availability of label data which is one of the most time-consuming requirements in supervised deep learning. In this paper, an advanced learning approach is proposed for geoscience where the machine observes the seismic interpretation activities and learns simultaneously as the interpretation progresses. Initial testing showed that through the proposed method along with transfer learning, machine learning performance is highly effective, and the machine accurately predicts features requiring minor post prediction filtering to be accepted as the optimal interpretation.

scholarly journals MAT-Net: Medial Axis Transform Network for 3D Object Recognition

2019 ◽  
Author(s):  
Jianwei Hu ◽  
Bin Wang ◽  
Lihui Qian ◽  
Yiling Pan ◽  
Xiaohu Guo ◽  
...  
Keyword(s):  
Deep Learning ◽  
Object Representation ◽  
Nearest Neighbor ◽  
Medial Axis ◽  
Fixed Number ◽  
Learning Performance ◽  
Medial Axis Transform ◽  
K Nearest Neighbor ◽  
Global Features ◽  
Local Feature Extraction

3D deep learning performance depends on object representation and local feature extraction. In this work, we present MAT-Net, a neural network which captures local and global features from the Medial Axis Transform (MAT). Different from K-Nearest-Neighbor method which extracts local features by a fixed number of neighbors, our MAT-Net exploits effective modules Group-MAT and Edge-Net to process topological structure. Experimental results illustrate that MAT-Net demonstrates competitive or better performance on 3D shape recognition than state-of-the-art methods, and prove that MAT representation has excellent capacity in 3D deep learning, even in the case of low resolution.

Unraveling Deep Learning Performance in Cross-Sensor Iris Recognition

2021 ◽  
pp. 93-104
Author(s):  
Meenakshi Choudhary ◽  
Vivek Tiwari ◽  
U. Venkanna
Keyword(s):  
Deep Learning ◽  
Iris Recognition ◽  
Learning Performance

Deep Learning Performance for Triage and Diagnosis

2020 ◽  
pp. 627-635
Author(s):  
Álvaro Iglesias-Puzas ◽  
Pablo Boixeda
Keyword(s):  
Deep Learning ◽  
Learning Performance

scholarly journals Meta-Learning Initializations for Low-Resource Drug Discovery

2020 ◽  
Author(s):  
Cuong Q. Nguyen ◽  
Constantine Kreatsoulas ◽  
Kim M. Branson
Keyword(s):  
Deep Learning ◽  
Drug Discovery ◽  
State Of The Art ◽  
Chemical Properties ◽  
Fine Tuning ◽  
Learning Performance ◽  
Low Resource ◽  
Average Improvement ◽  
Meta Learning ◽  
In Silico Models

Building in silico models to predict chemical properties and activities is a crucial step in drug discovery. However, drug discovery projects are often characterized by limited labeled data, hindering the applications of deep learning in this setting. Meanwhile advances in meta-learning have enabled state-of-the-art performances in few-shot learning benchmarks, naturally prompting the question: Can meta-learning improve deep learning performance in low-resource drug discovery projects? In this work, we assess the efficiency of the Model-Agnostic Meta-Learning (MAML) algorithm – along with its variants FO-MAML and ANIL – at learning to predict chemical properties and activities. Using the ChEMBL20 dataset to emulate low-resource settings, our benchmark shows that meta-initializations perform comparably to or outperform multi-task pre-training baselines on 16 out of 20 in-distribution tasks and on all out-of-distribution tasks, providing an average improvement in AUPRC of 7.2% and 14.9% respectively. Finally, we observe that meta-initializations consistently result in the best performing models across fine-tuning sets with k ∈ {16, 32, 64, 128, 256} instances.<br>

Sign in / Sign up

Export Citation Format

Share Document