BiHMP-GAN: Bidirectional 3D Human Motion Prediction GAN

Kundu, Jogendra Nath; Gor, Maharshi; Babu, R. Venkatesh

doi:10.48550/arxiv.1812.02591

Cited by 2 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While we present some simple experimental results to validate our theory, our focus here is on the theoretical ideas; the important and challenging problem of designing good generative models to implement the proposed defense for general AI classification systems is deferred to future work. Interestingly, while our theory is novel, other researchers have recently developed defenses for AI classifiers against adversarial attacks that are consistent with our proposed approach [11], [12]. An AI classifier can be defined as a system that takes a high-dimensional vector as input and maps it to a discrete set of labels.…”

Section: Introductionmentioning

confidence: 66%

An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers

Xie

et al. 2019

2019 IEEE International Symposium on Information Theory (ISIT)

View full text Add to dashboard Cite

We present a simple hypothesis about a compression property of artificial intelligence (AI) classifiers and present theoretical arguments to show that this hypothesis successfully accounts for the observed fragility of AI classifiers to small adversarial perturbations. We also propose a new method for detecting when small input perturbations cause classifier errors, and show theoretical guarantees for the performance of this detection method. We present experimental results with a voice recognition system to demonstrate this method. The ideas in this paper are motivated by a simple analogy between AI classifiers and the standard Shannon model of a communication system. 1

show abstract

Section: Introductionmentioning

confidence: 66%

An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers

Xie

et al. 2019

2019 IEEE International Symposium on Information Theory (ISIT)

View full text Add to dashboard Cite

show abstract

“…Recently, several attempts have been made at modeling the stochastic nature of human motion [33,3,31,19,21]. These methods rely on sampling a random vector that is then combined with an encoding of the observed pose sequence.…”

Section: Mean Perturbed Hidden State Diversitymentioning

confidence: 99%

“…By relying on different random vectors at each time step, however, this strategy is prone to generating discontinuous motions. To overcome this, [19] makes use of a single random vector to generate the entire sequence. This vector is both employed to alter the initialization of the decoder and concatenated with a pose embedding at each iteration of the RNN.…”

Section: Related Workmentioning

confidence: 99%

“…Human motion prediction aims to forecast the sequence of future poses of a person given past observations of such poses. To achieve this, existing methods typically rely on recurrent neural networks (RNNs) that encode the person's motion [22,11,31,19,3,25,26]. While they predict reasonable motions, RNNs are deterministic models and thus cannot account for the highly stochastic nature of human motion; given the beginning of a sequence, multiple, diverse futures are plausible.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Variations in Human Motion via Mix-and-Match Perturbation

Aliakbarian,

Saleh,

Salzmann

et al. 2019

Preprint

View full text Add to dashboard Cite

Human motion prediction is a stochastic process: Given an observed sequence of poses, multiple future motions are plausible. Existing approaches to modeling this stochasticity typically combine a random noise vector with information about the previous poses. This combination, however, is done in a deterministic manner, which gives the network the flexibility to learn to ignore the random noise. In this paper, we introduce an approach to stochastically combine the root of variations with previous pose information, which forces the model to take the noise into account. We exploit this idea for motion prediction by incorporating it into a recurrent encoder-decoder network with a conditional variational autoencoder block that learns to exploit the perturbations. Our experiments demonstrate that our model yields high-quality pose sequences that are much more diverse than those from state-of-the-art stochastic motion prediction techniques.

show abstract

BiHMP-GAN: Bidirectional 3D Human Motion Prediction GAN

Cited by 2 publications

References 0 publications

An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers

An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers

Learning Variations in Human Motion via Mix-and-Match Perturbation

Contact Info

Product

Resources

About