scholarly journals Regional Tree Regularization for Interpretability in Deep Neural Networks

2020 ◽  
Vol 34 (04) ◽  
pp. 6413-6421
Author(s):  
Mike Wu ◽  
Sonali Parbhoo ◽  
Michael Hughes ◽  
Ryan Kindle ◽  
Leo Celi ◽  
...  
Keyword(s):  
Neural Networks ◽  
Decision Tree ◽  
Decision Trees ◽  
Deep Neural Networks ◽  
Healthcare Applications ◽  
Input Space ◽  
Safety Critical ◽  
Single Tree ◽  
Deep Model

The lack of interpretability remains a barrier to adopting deep neural networks across many safety-critical domains. Tree regularization was recently proposed to encourage a deep neural network's decisions to resemble those of a globally compact, axis-aligned decision tree. However, it is often unreasonable to expect a single tree to predict well across all possible inputs. In practice, doing so could lead to neither interpretable nor performant optima. To address this issue, we propose regional tree regularization – a method that encourages a deep model to be well-approximated by several separate decision trees specific to predefined regions of the input space. Across many datasets, including two healthcare applications, we show our approach delivers simpler explanations than other regularization schemes without compromising accuracy. Specifically, our regional regularizer finds many more “desirable” optima compared to global analogues.

scholarly journals SADA: Semantic Adversarial Diagnostic Attacks for Autonomous Applications

2020 ◽  
Vol 34 (07) ◽  
pp. 10901-10908 ◽  
Author(s):  
Abdullah Hamdi ◽  
Matthias Mueller ◽  
Bernard Ghanem
Keyword(s):  
Neural Networks ◽  
Recent Work ◽  
Autonomous Navigation ◽  
General Framework ◽  
Deep Neural Networks ◽  
Autonomous Driving ◽  
Black Box ◽  
Semantic Meaning ◽  
Safety Critical ◽  
Adversarial Attack

One major factor impeding more widespread adoption of deep neural networks (DNNs) is their lack of robustness, which is essential for safety-critical applications such as autonomous driving. This has motivated much recent work on adversarial attacks for DNNs, which mostly focus on pixel-level perturbations void of semantic meaning. In contrast, we present a general framework for adversarial attacks on trained agents, which covers semantic perturbations to the environment of the agent performing the task as well as pixel-level attacks. To do this, we re-frame the adversarial attack problem as learning a distribution of parameters that always fools the agent. In the semantic case, our proposed adversary (denoted as BBGAN) is trained to sample parameters that describe the environment with which the black-box agent interacts, such that the agent performs its dedicated task poorly in this environment. We apply BBGAN on three different tasks, primarily targeting aspects of autonomous navigation: object detection, self-driving, and autonomous UAV racing. On these tasks, BBGAN can generate failure cases that consistently fool a trained agent.

scholarly journals Self-Supervised Learning for Generalizable Out-of-Distribution Detection

2020 ◽  
Vol 34 (04) ◽  
pp. 5216-5223 ◽  
Author(s):  
Sina Mohseni ◽  
Mandar Pitale ◽  
JBS Yadawa ◽  
Zhangyang Wang
Keyword(s):  
Neural Networks ◽  
Autonomous Vehicles ◽  
Deep Neural Networks ◽  
State Of The Art ◽  
Feature Learning ◽  
Detection Methods ◽  
Training Set ◽  
Safety Critical ◽  
Multiple Image ◽  
A New Technique

The real-world deployment of Deep Neural Networks (DNNs) in safety-critical applications such as autonomous vehicles needs to address a variety of DNNs' vulnerabilities, one of which being detecting and rejecting out-of-distribution outliers that might result in unpredictable fatal errors. We propose a new technique relying on self-supervision for generalizable out-of-distribution (OOD) feature learning and rejecting those samples at the inference time. Our technique does not need to pre-know the distribution of targeted OOD samples and incur no extra overheads compared to other methods. We perform multiple image classification experiments and observe our technique to perform favorably against state-of-the-art OOD detection methods. Interestingly, we witness that our method also reduces in-distribution classification risk via rejecting samples near the boundaries of the training set distribution.

scholarly journals A New Aggregated Attribute Values Match Technique for Improving the Quality of Probability Estimated Decision Trees

2020 ◽  
Vol 9 (7) ◽  
pp. 446-452
Keyword(s):  
Decision Tree ◽  
Decision Trees ◽  
Area Under The Curve ◽  
Probability Estimation ◽  
Evaluation Technique ◽  
Tree Model ◽  
Match Technique ◽  
Single Tree ◽  
Model Technique

Probability estimations of decision trees may not be useful directly because their poor probability estimations but the best probability estimations are desired in many useful applications. Many techniques have been proposed for obtaining good probability estimations of decision trees. Two such optical techniques are identified and the first one is single tree based aggregation of mismatched attribute values of instances. The second one is bagging technique but it is costly and less comprehensible. So, in this paper a single aggregated probability estimation decision tree model technique is proposed for improving the performance of probability estimations of decision trees and the performance of new technique is evaluated using area under the curve (AUC) evaluation technique. The proposed technique computes aggregate scores based on matched attribute values of test tuples.

scholarly journals Analyzing the instructions vulnerability of dense convolutional network on GPUS

2021 ◽  
Vol 11 (5) ◽  
pp. 4481
Author(s):  
Khalid Adam ◽  
Izzeldin I. Mohd ◽  
Younis Ibrahim
Keyword(s):  
Graphics Processing Units ◽  
Deep Neural Networks ◽  
Fault Injection ◽  
Soft Errors ◽  
Small Error ◽  
Mitigation Strategy ◽  
Healthcare Applications ◽  
Convolutional Network ◽  
Safety Critical ◽  
Graphics Processing

Recently, deep neural networks (DNNs) have been increasingly deployed in various healthcare applications, which are considered safety-critical applications. Thus, the reliability of these DNN models should be remarkably high, because even a small error in healthcare applications can lead to injury or death. Due to the high computations of the DNN models, DNNs are often executed on the graphics processing units (GPUs). However, the GPUs have been reportedly impacted by soft errors, which are extremely serious issues in the healthcare applications. In this paper, we show how the fault injection can provide a deeper understanding of DenseNet201 model instructions vulnerability on the GPU. Then, we analyze vulnerable instructions of the DenseNet201 on the GPU. Our results show that the most significant vulnerable instructions against soft errors PR, STORE, FADD, FFMA, SETP and LD can be reduced from 4.42% to 0.14% of injected faults, after we applied our mitigation strategy.

scholarly journals Optimizing for Interpretability in Deep Neural Networks with Tree Regularization

2021 ◽  
Vol 72 ◽  
pp. 1-37
Author(s):  
Mike Wu ◽  
Sonali Parbhoo ◽  
Michael C. Hughes ◽  
Volker Roth ◽  
Finale Doshi-Velez
Keyword(s):  
Neural Networks ◽  
Predictive Power ◽  
Deep Neural Networks ◽  
Large Body ◽  
Black Box ◽  
New Family ◽  
Deep Model ◽  
Real World Applications ◽  
Adversarial Examples ◽  
Key Barrier

Deep models have advanced prediction in many domains, but their lack of interpretability  remains a key barrier to the adoption in many real world applications. There exists a large  body of work aiming to help humans understand these black box functions to varying levels  of granularity – for example, through distillation, gradients, or adversarial examples. These  methods however, all tackle interpretability as a separate process after training. In this  work, we take a different approach and explicitly regularize deep models so that they are  well-approximated by processes that humans can step through in little time. Specifically,  we train several families of deep neural networks to resemble compact, axis-aligned decision  trees without significant compromises in accuracy. The resulting axis-aligned decision  functions uniquely make tree regularized models easy for humans to interpret. Moreover,  for situations in which a single, global tree is a poor estimator, we introduce a regional tree regularizer that encourages the deep model to resemble a compact, axis-aligned decision  tree in predefined, human-interpretable contexts. Using intuitive toy examples, benchmark  image datasets, and medical tasks for patients in critical care and with HIV, we demonstrate  that this new family of tree regularizers yield models that are easier for humans to simulate  than L1 or L2 penalties without sacrificing predictive power. 

scholarly journals SpecAugment++: A Hidden Space Data Augmentation Method for AcousticScene Classification

2021 ◽  
Author(s):  
Helin Wang ◽  
Yuexian Zou ◽  
Wenwu Wang
Keyword(s):  
Neural Networks ◽  
Deep Neural Networks ◽  
Data Augmentation ◽  
Input Space ◽  
Space Data ◽  
Time Frames

In this paper, we present SpecAugment++, a novel data aug-mentation method for deep neural networks based acousticscene classification (ASC). Different from other popular dataaugmentation methods such as SpecAugment and mixup thatonly work on the input space, SpecAugment++ is applied toboth the input space and the hidden space of the deep neuralnetworks to enhance the input and the intermediate feature rep-resentations. For an intermediate hidden state, the augmentationtechniques consist of masking blocks of frequency channels andmasking blocks of time frames, which improve generalizationby enabling a model to attend not only to the most discrimina-tive parts of the feature, but also the entire parts. Apart fromusing zeros for masking, we also examine two approaches formasking based on the use of other samples within the mini-batch, which helps introduce noises to the networks to makethem more discriminative for classification. The experimentalresults on the DCASE 2018 Task1 dataset and DCASE 2019Task1 dataset show that our proposed method can obtain3.6%and4.7%accuracy gains over a strong baseline without aug-mentation (i.e.CP-ResNet) respectively, and outperforms otherprevious data augmentation methods.

scholarly journals Towards Understanding the Risks of Gradient Inversion in Federated Learning

2021 ◽  
Author(s):  
Ali Hatamizadeh ◽  
Hongxu Yin ◽  
Pavlo Molchanov ◽  
Andriy Myronenko ◽  
Wenqi Li ◽  
...  
Keyword(s):  
Neural Networks ◽  
Data Privacy ◽  
Deep Neural Networks ◽  
Differential Privacy ◽  
Use Cases ◽  
Training Data ◽  
Model Accuracy ◽  
Healthcare Applications ◽  
Raw Data ◽  
Collaborative Training

Abstract Federated learning (FL) allows the collaborative training of AI models without needing to share raw data. This capability makes it especially interesting for healthcare applications where patient and data privacy is of utmost concern. However, recent works on the inversion of deep neural networks from model gradients raised concerns about the security of FL in preventing the leakage of training data. In this work, we show that these attacks presented in the literature are impractical in real FL use-cases and provide a new baseline attack that works for more realistic scenarios where the clients’ training involves updating the Batch Normalization (BN) statistics. Furthermore, we present new ways to measure and visualize potential data leakage in FL. Our work is a step towards establishing reproducible methods of measuring data leakage in FL and could help determine the optimal tradeoffs between privacy-preserving techniques, such as differential privacy, and model accuracy based on quantifiable metrics.

scholarly journals Safety Analysis of Deep Neural Networks

2021 ◽  
Author(s):  
Dario Guidotti
Keyword(s):  
Machine Learning ◽  
Neural Networks ◽  
Computer Science ◽  
Deep Neural Networks ◽  
Safety Analysis ◽  
Learning Models ◽  
Safety Critical ◽  
Machine Learning Models

Deep Neural Networks (DNNs) are popular machine learning models which have found successful application in many different domains across computer science. Nevertheless, providing formal guarantees on the behaviour of neural networks is hard and therefore their reliability in safety-critical domains is still a concern. Verification and repair emerged as promising solutions to address this issue. In the following, I will present some of my recent efforts in this area.

scholarly journals Integrating ECG Monitoring and Classification via IoT and Deep Neural Networks

Biosensors ◽  
2021 ◽  
Vol 11 (6) ◽  
pp. 188
Author(s):  
Li-Ren Yeh ◽  
Wei-Chin Chen ◽  
Hua-Yan Chan ◽  
Nan-Han Lu ◽  
Chi-Yuan Wang ◽  
...  
Keyword(s):  
Neural Networks ◽  
Real Time ◽  
Deep Neural Networks ◽  
Network Models ◽  
Kappa Statistics ◽  
Neural Network Models ◽  
Ecg Signals ◽  
St Depression ◽  
Deep Model ◽  
St Elevation

Anesthesia assessment is most important during surgery. Anesthesiologists use electrocardiogram (ECG) signals to assess the patient’s condition and give appropriate medications. However, it is not easy to interpret the ECG signals. Even physicians with more than 10 years of clinical experience may still misjudge. Therefore, this study uses convolutional neural networks to classify ECG image types to assist in anesthesia assessment. The research uses Internet of Things (IoT) technology to develop ECG signal measurement prototypes. At the same time, it classifies signal types through deep neural networks, divided into QRS widening, sinus rhythm, ST depression, and ST elevation. Three models, ResNet, AlexNet, and SqueezeNet, are developed with 50% of the training set and test set. Finally, the accuracy and kappa statistics of ResNet, AlexNet, and SqueezeNet in ECG waveform classification were (0.97, 0.96), (0.96, 0.95), and (0.75, 0.67), respectively. This research shows that it is feasible to measure ECG in real time through IoT and then distinguish four types through deep neural network models. In the future, more types of ECG images will be added, which can improve the real-time classification practicality of the deep model.

Sign in / Sign up

Export Citation Format

Share Document