scholarly journals Integer-Only CNNs with 4 Bit Weights and Bit-Shift Quantization Scales at Full-Precision Accuracy

Electronics ◽  
2021 ◽  
Vol 10 (22) ◽  
pp. 2823
Author(s):  
Maarten Vandersteegen ◽  
Kristof Van Beeck ◽  
Toon Goedemé
Keyword(s):  
Neural Networks ◽  
High Precision ◽  
State Of The Art ◽  
Lookup Table ◽  
Memory Bandwidth ◽  
Quantization Scheme ◽  
Hardware Cost ◽  
Channel Quantization ◽  
Diverse Data ◽  
Hardware Platforms

Quantization of neural networks has been one of the most popular techniques to compress models for embedded (IoT) hardware platforms with highly constrained latency, storage, memory-bandwidth, and energy specifications. Limiting the number of bits per weight and activation has been the main focus in the literature. To avoid major degradation of accuracy, common quantization methods introduce additional scale factors to adapt the quantized values to the diverse data ranges, present in full-precision (floating-point) neural networks. These scales are usually kept in high precision, requiring the target compute engine to support a few high-precision multiplications, which is not desirable due to the larger hardware cost. Little effort has yet been invested in trying to avoid high-precision multipliers altogether, especially in combination with 4 bit weights. This work proposes a new quantization scheme, based on power-of-two quantization scales, that works on-par compared to uniform per-channel quantization with full-precision 32 bit quantization scales when using only 4 bit weights. This is done through the addition of a low-precision lookup-table that translates stored 4 bit weights into nonuniformly distributed 8 bit weights for internal computation. All our quantized ImageNet CNNs achieved or even exceeded the Top-1 accuracy of their full-precision counterparts, with ResNet18 exceeding its full-precision model by 0.35%. Our MobileNetV2 model achieved state-of-the-art performance with only a slight drop in accuracy of 0.51%.

scholarly journals HLHLp: Quantized Neural Networks Training for Reaching Flat Minima in Loss Surface

2020 ◽  
Vol 34 (04) ◽  
pp. 5784-5791
Author(s):  
Sungho Shin ◽  
Jinhwan Park ◽  
Yoonho Boo ◽  
Wonyong Sung
Keyword(s):  
Neural Networks ◽  
Deep Neural Networks ◽  
State Of The Art ◽  
Fine Tuning ◽  
Quantization Noise ◽  
Quantization Scheme ◽  
Performance Improvements ◽  
Training Scheme ◽  
Network Training ◽  
Training Technique

Quantization of deep neural networks is extremely essential for efficient implementations. Low-precision networks are typically designed to represent original floating-point counterparts with high fidelity, and several elaborate quantization algorithms have been developed. We propose a novel training scheme for quantized neural networks to reach flat minima in the loss surface with the aid of quantization noise. The proposed training scheme employs high-low-high-low precision in an alternating manner for network training. The learning rate is also abruptly changed at each stage for coarse- or fine-tuning. With the proposed training technique, we show quite good performance improvements for convolutional neural networks when compared to the previous fine-tuning based quantization scheme. We achieve the state-of-the-art results for recurrent neural network based language modeling with 2-bit weight and activation.

scholarly journals Designing Automated Deployment Strategies of Face Recognition Solutions in Heterogeneous IoT Platforms

Information ◽  
2021 ◽  
Vol 12 (12) ◽  
pp. 532
Author(s):  
Unai Elordi ◽  
Chiara Lunerti ◽  
Luis Unzueta ◽  
Jon Goenetxea ◽  
Nerea Aranjuelo ◽  
...  
Keyword(s):  
Neural Networks ◽  
Face Recognition ◽  
Deep Neural Networks ◽  
State Of The Art ◽  
User Interaction ◽  
Biometric Data ◽  
Optimal Deployment ◽  
Iot Platforms ◽  
Iot Devices ◽  
Hardware Platforms

In this paper, we tackle the problem of deploying face recognition (FR) solutions in heterogeneous Internet of Things (IoT) platforms. The main challenges are the optimal deployment of deep neural networks (DNNs) in the high variety of IoT devices (e.g., robots, tablets, smartphones, etc.), the secure management of biometric data while respecting the users’ privacy, and the design of appropriate user interaction with facial verification mechanisms for all kinds of users. We analyze different approaches to solving all these challenges and propose a knowledge-driven methodology for the automated deployment of DNN-based FR solutions in IoT devices, with the secure management of biometric data, and real-time feedback for improved interaction. We provide some practical examples and experimental results with state-of-the-art DNNs for FR in Intel’s and NVIDIA’s hardware platforms as IoT devices.

scholarly journals Dynamic Hypergraph Neural Networks

2019 ◽  
Author(s):  
Jianwen Jiang ◽  
Yuxuan Wei ◽  
Yifan Feng ◽  
Jingxuan Cao ◽  
Yue Gao
Keyword(s):  
Neural Networks ◽  
Deep Learning ◽  
Prior Knowledge ◽  
State Of The Art ◽  
Citation Network ◽  
High Order ◽  
Learning Methods ◽  
Inherent Structure ◽  
Two Phases ◽  
Diverse Data

In recent years, graph/hypergraph-based deep learning methods have attracted much attention from researchers. These deep learning methods take graph/hypergraph structure as prior knowledge in the model. However, hidden and important relations are not directly represented in the inherent structure. To tackle this issue, we propose a dynamic hypergraph neural networks framework (DHGNN), which is composed of the stacked layers of two modules: dynamic hypergraph construction (DHG) and hypergrpah convolution (HGC). Considering initially constructed hypergraph is probably not a suitable representation for data, the DHG module dynamically updates hypergraph structure on each layer. Then hypergraph convolution is introduced to encode high-order data relations in a hypergraph structure. The HGC module includes two phases: vertex convolution and hyperedge convolution, which are designed to aggregate feature among vertices and hyperedges, respectively. We have evaluated our method on standard datasets, the Cora citation network and Microblog dataset. Our method outperforms state-of-the-art methods. More experiments are conducted to demonstrate the effectiveness and robustness of our method to diverse data distributions.

AutoLinker: Automatic Fragment Linking with Deep Conditional Transformer Neural Networks

2020 ◽  
Author(s):  
Yuyao Yang ◽  
Shuangjia Zheng ◽  
Shimin Su ◽  
Jun Xu ◽  
Hongming Chen
Keyword(s):  
Neural Networks ◽  
Drug Discovery ◽  
Drug Design ◽  
State Of The Art ◽  
The State ◽  
Generative Model ◽  
Lead Optimization ◽  
Scaffold Hopping ◽  
Reference Models ◽  
Lead Generation

Fragment based drug design represents a promising drug discovery paradigm complimentary to the traditional HTS based lead generation strategy. How to link fragment structures to increase compound affinity is remaining a challenge task in this paradigm. Hereby a novel deep generative model (AutoLinker) for linking fragments is developed with the potential for applying in the fragment-based lead generation scenario. The state-of-the-art transformer architecture was employed to learn the linker grammar and generate novel linker. Our results show that, given starting fragments and user customized linker constraints, our AutoLinker model can design abundant drug-like molecules fulfilling these constraints and its performance was superior to other reference models. Moreover, several examples were showcased that AutoLinker can be useful tools for carrying out drug design tasks such as fragment linking, lead optimization and scaffold hopping.

Levenshtein Augmentation Improves Performance of SMILES Based Deep-Learning Synthesis Prediction

2020 ◽  
Author(s):  
Dean Sumner ◽  
Jiazhen He ◽  
Amol Thakkar ◽  
Ola Engkvist ◽  
Esben Jannik Bjerrum
Keyword(s):  
Neural Networks ◽  
Pattern Recognition ◽  
Deep Learning ◽  
Recurrent Neural Networks ◽  
Data Augmentation ◽  
State Of The Art ◽  
Sequence Similarity ◽  
Learning Models ◽  
Underlying Network

<p>SMILES randomization, a form of data augmentation, has previously been shown to increase the performance of deep learning models compared to non-augmented baselines. Here, we propose a novel data augmentation method we call “Levenshtein augmentation” which considers local SMILES sub-sequence similarity between reactants and their respective products when creating training pairs. The performance of Levenshtein augmentation was tested using two state of the art models - transformer and sequence-to-sequence based recurrent neural networks with attention. Levenshtein augmentation demonstrated an increase performance over non-augmented, and conventionally SMILES randomization augmented data when used for training of baseline models. Furthermore, Levenshtein augmentation seemingly results in what we define as <i>attentional gain </i>– an enhancement in the pattern recognition capabilities of the underlying network to molecular motifs.</p>

scholarly journals Using spatial-temporal ensembles of convolutional neural networks for lumen segmentation in ureteroscopy

2021 ◽  
Author(s):  
Jorge F. Lazo ◽  
Aldo Marzullo ◽  
Sara Moccia ◽  
Michele Catellani ◽  
Benoit Rosa ◽  
...  
Keyword(s):  
Neural Networks ◽  
Convolutional Neural Networks ◽  
State Of The Art ◽  
Automatic Segmentation ◽  
Temporal Information ◽  
Invasive Technique ◽  
Dice Similarity Coefficient ◽  
Specular Reflections ◽  
Lumen Segmentation ◽  
Previous State

Abstract Purpose Ureteroscopy is an efficient endoscopic minimally invasive technique for the diagnosis and treatment of upper tract urothelial carcinoma. During ureteroscopy, the automatic segmentation of the hollow lumen is of primary importance, since it indicates the path that the endoscope should follow. In order to obtain an accurate segmentation of the hollow lumen, this paper presents an automatic method based on convolutional neural networks (CNNs). Methods The proposed method is based on an ensemble of 4 parallel CNNs to simultaneously process single and multi-frame information. Of these, two architectures are taken as core-models, namely U-Net based in residual blocks ($$m_1$$ m 1 ) and Mask-RCNN ($$m_2$$ m 2 ), which are fed with single still-frames I(t). The other two models ($$M_1$$ M 1 , $$M_2$$ M 2 ) are modifications of the former ones consisting on the addition of a stage which makes use of 3D convolutions to process temporal information. $$M_1$$ M 1 , $$M_2$$ M 2 are fed with triplets of frames ($$I(t-1)$$ I ( t - 1 ) , I(t), $$I(t+1)$$ I ( t + 1 ) ) to produce the segmentation for I(t). Results The proposed method was evaluated using a custom dataset of 11 videos (2673 frames) which were collected and manually annotated from 6 patients. We obtain a Dice similarity coefficient of 0.80, outperforming previous state-of-the-art methods. Conclusion The obtained results show that spatial-temporal information can be effectively exploited by the ensemble model to improve hollow lumen segmentation in ureteroscopic images. The method is effective also in the presence of poor visibility, occasional bleeding, or specular reflections.

scholarly journals Improving Semi-Supervised Learning for Audio Classification with FixMatch

Electronics ◽  
2021 ◽  
Vol 10 (15) ◽  
pp. 1807
Author(s):  
Sascha Grollmisch ◽  
Estefanía Cano
Keyword(s):  
Neural Networks ◽  
Supervised Learning ◽  
Transfer Learning ◽  
Data Transfer ◽  
State Of The Art ◽  
Training Data ◽  
Audio Classification ◽  
Image Domain ◽  
Full Dataset ◽  
Audio Data

Including unlabeled data in the training process of neural networks using Semi-Supervised Learning (SSL) has shown impressive results in the image domain, where state-of-the-art results were obtained with only a fraction of the labeled data. The commonality between recent SSL methods is that they strongly rely on the augmentation of unannotated data. This is vastly unexplored for audio data. In this work, SSL using the state-of-the-art FixMatch approach is evaluated on three audio classification tasks, including music, industrial sounds, and acoustic scenes. The performance of FixMatch is compared to Convolutional Neural Networks (CNN) trained from scratch, Transfer Learning, and SSL using the Mean Teacher approach. Additionally, a simple yet effective approach for selecting suitable augmentation methods for FixMatch is introduced. FixMatch with the proposed modifications always outperformed Mean Teacher and the CNNs trained from scratch. For the industrial sounds and music datasets, the CNN baseline performance using the full dataset was reached with less than 5% of the initial training data, demonstrating the potential of recent SSL methods for audio data. Transfer Learning outperformed FixMatch only for the most challenging dataset from acoustic scene classification, showing that there is still room for improvement.

scholarly journals SketchGNN: Semantic Sketch Segmentation with Graph Neural Networks

2021 ◽  
Vol 40 (3) ◽  
pp. 1-13
Author(s):  
Lumin Yang ◽  
Jiajie Zhuang ◽  
Hongbo Fu ◽  
Xiangzhi Wei ◽  
Kun Zhou ◽  
...  
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Network Architecture ◽  
Large Scale ◽  
State Of The Art ◽  
Semantic Segmentation ◽  
Structure Information ◽  
Graph Neural Networks ◽  
Node Labels ◽  
Point Level

We introduce SketchGNN , a convolutional graph neural network for semantic segmentation and labeling of freehand vector sketches. We treat an input stroke-based sketch as a graph with nodes representing the sampled points along input strokes and edges encoding the stroke structure information. To predict the per-node labels, our SketchGNN uses graph convolution and a static-dynamic branching network architecture to extract the features at three levels, i.e., point-level, stroke-level, and sketch-level. SketchGNN significantly improves the accuracy of the state-of-the-art methods for semantic sketch segmentation (by 11.2% in the pixel-based metric and 18.2% in the component-based metric over a large-scale challenging SPG dataset) and has magnitudes fewer parameters than both image-based and sequence-based methods.

scholarly journals A Review of Algorithms and Hardware Implementations for Spiking Neural Networks

2021 ◽  
Vol 11 (2) ◽  
pp. 23
Author(s):  
Duy-Anh Nguyen ◽  
Xuan-Tu Tran ◽  
Francesca Iacopi
Keyword(s):  
Neural Networks ◽  
Computational Cost ◽  
Superior Performance ◽  
Training Algorithms ◽  
Current State ◽  
Hardware Implementations ◽  
Neuromorphic Hardware ◽  
High Level ◽  
Event Based ◽  
Hardware Platforms

Deep Learning (DL) has contributed to the success of many applications in recent years. The applications range from simple ones such as recognizing tiny images or simple speech patterns to ones with a high level of complexity such as playing the game of Go. However, this superior performance comes at a high computational cost, which made porting DL applications to conventional hardware platforms a challenging task. Many approaches have been investigated, and Spiking Neural Network (SNN) is one of the promising candidates. SNN is the third generation of Artificial Neural Networks (ANNs), where each neuron in the network uses discrete spikes to communicate in an event-based manner. SNNs have the potential advantage of achieving better energy efficiency than their ANN counterparts. While generally there will be a loss of accuracy on SNN models, new algorithms have helped to close the accuracy gap. For hardware implementations, SNNs have attracted much attention in the neuromorphic hardware research community. In this work, we review the basic background of SNNs, the current state and challenges of the training algorithms for SNNs and the current implementations of SNNs on various hardware platforms.

Sign in / Sign up

Export Citation Format

Share Document