Generalization Error in Deep Learning

Jakubovitz, Daniel; Giryes, Raja; Rodrigues, Miguel R. D.

doi:10.1007/978-3-319-73074-5_5

Cited by 80 publications

(41 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the one hand, this creates a safety risk if the model is exposed to benign inputs which sufficiently differ from those seen during training, such that the model is unable to generalize to these new inputs (Novak et al, 2018 ; Jakubovitz et al, 2019 ). The probability of this happening depends on many factors, including the model, the algorithm used and especially the quality of the training data (Chung et al, 2018 ; Zahavy et al, 2018 ).…”

Section: Key Factors Underlying Ai-specific Vulnerabilitiesmentioning

confidence: 99%

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

2020

View full text Add to dashboard Cite

This article deals with the IT security of connectionist artificial intelligence (AI) applications, focusing on threats to integrity, one of the three IT security goals. Such threats are for instance most relevant in prominent AI computer vision applications. In order to present a holistic view on the IT security goal integrity, many additional aspects, such as interpretability, robustness and documentation are taken into account. A comprehensive list of threats and possible mitigations is presented by reviewing the state-of-the-art literature. AI-specific vulnerabilities, such as adversarial attacks and poisoning attacks are discussed in detail, together with key factors underlying them. Additionally and in contrast to former reviews, the whole AI life cycle is analyzed with respect to vulnerabilities, including the planning, data acquisition, training, evaluation and operation phases. The discussion of mitigations is likewise not restricted to the level of the AI system itself but rather advocates viewing AI systems in the context of their life cycles and their embeddings in larger IT infrastructures and hardware devices. Based on this and the observation that adaptive attackers may circumvent any single published AI-specific defense to date, the article concludes that single protective measures are not sufficient but rather multiple measures on different levels have to be combined to achieve a minimum level of IT security for AI applications.

show abstract

Section: Key Factors Underlying Ai-specific Vulnerabilitiesmentioning

confidence: 99%

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

2020

View full text Add to dashboard Cite

show abstract

“…Ensuring generalization of deep learning models is a challenging and on-going research area [50]. It becomes more challenging when the training data and the testing data come from different data distributions.…”

Section: The Importance Of Generating Realistic Simulation Datamentioning

confidence: 99%

SIFNet: Electromagnetic Source Imaging Framework Using Deep Neural Networks

Sun

Sohrabpour

et al. 2020

Preprint

View full text Add to dashboard Cite

AbstractElectroencephalography (EEG) and magnetoencephalography (MEG) are used to measure brain activity, noninvasively, and are useful tools for brain research and clinical management of brain disorders. Tremendous effort has been made in solving the inverse source imaging problem from EEG/MEG measurements. This is a challenging ill-posed problem, since the number of measurements is much smaller than the number of possible sources in the brain. Various methods have been developed to estimate underlying brain sources from noninvasive EEG/MEG as this can offer insight about the underlying brain electrical activity with significantly improved spatial resolution. In this work, we propose a novel data-driven Source Imaging Framework using deep learning neural networks (SIFNet), where (1) a simulation pipeline is designed to model realistic brain activation and EEG/MEG signals to train generalizable neural networks, (2) and a residual convolutional neural network is trained using the simulated data, capable of estimating source distributions from EEG/MEG recordings. The performance of our proposed SIFNet approach is evaluated in a series of computer simulations, which indicates the excellent performance of SIFNet outperforming conventional weighted minimum norm algorithms that were tested in this work. The SIFNet is further tested by analyzing interictal EEG data recorded in a clinical setting from a focal epilepsy patient. The results of this clinical data analysis indicate accurate localization of epileptogenic activity as validated by the epileptogenic zone clinically determined in this patient. In sum, the proposed SIFNet approach promises to offer an alternative solution to the EEG/MEG inverse source imaging problem, shows promising signs of being robust against measurement noise, and is easy to implement, therefore, being translatable to clinical practice.

show abstract

“…The generalization error of a machine learning model is defined as the difference between the empirical loss of the training set and the expected loss of test set [36]. This measure represents the ability of the trained model to generalize well from the learning data to new unseen data, thereby being able to extrapolate from training data to new test data.…”

Section: A Stability Analysismentioning

confidence: 99%

Deep Model Reference Adaptive Control

Joshi

Chowdhary

2019

2019 IEEE 58th Conference on Decision and Control (CDC)

View full text Add to dashboard Cite

We present a new neuroadaptive architecture: Deep Neural Network based Model Reference Adaptive Control (DMRAC). Our architecture utilizes the power of deep neural network representations for modeling significant nonlinearities while marrying it with the boundedness guarantees that characterize MRAC based controllers. We demonstrate through simulations and analysis that DMRAC can subsume previously studied learning based MRAC methods, such as concurrent learning and GP-MRAC. This makes DMRAC a highly powerful architecture for high-performance control of nonlinear systems with long-term learning properties.

show abstract

Generalization Error in Deep Learning

Cited by 80 publications

References 29 publications

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

Vulnerabilities of Connectionist AI Applications: Evaluation and Defense

SIFNet: Electromagnetic Source Imaging Framework Using Deep Neural Networks

Deep Model Reference Adaptive Control

Contact Info

Product

Resources

About