Particle Filter Recurrent Neural Networks

Ma, Xiao; Karkus, Péter; Hsu, David; Lee, Wee Sun

doi:10.1609/aaai.v34i04.5952

Cited by 56 publications

(46 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This shows that the ability to detect input values outside of the training distribution would be a valuable addition to current DFs. Finally, it would be interesting to compare learning in DFs to similar variational methods such as the ones introduced by Karl et al (2017); Fraccaro et al (2017); Le et al (2018) or the model-free PF-RNNs introduced by Ma et al (2020).…”

Section: Discussionmentioning

confidence: 99%

“…Integrating algorithmic structure into learning methods has been studied for many robotic problems, including state estimation (Haarnoja et al 2016;Jonschkowski and Brock 2016;Jonschkowski et al 2018;Karkus et al 2018;Ma et al 2020), planning (Tamar et al 2016;Karkus et al 2017;Oh et al 2017;Farquhar et al 2018;Guez et al 2018) and control (Donti et al 2017;Okada et al 2017;Amos et al 2018;Pereira et al 2018;Holl et al 2020). Most notably, Karkus et al (2019) combine multiple differentiable algorithms into an end-to-end trainable "Differentiable Algorithm Network" to address the complete task of navigating to a goal in a previously unseen environment using visual observations.…”

Section: Combining Learning and Algorithmsmentioning

confidence: 99%

See 1 more Smart Citation

How to train your differentiable filter

2021

View full text Add to dashboard Cite

In many robotic applications, it is crucial to maintain a belief about the state of a system, which serves as input for planning and decision making and provides feedback during task execution. Bayesian Filtering algorithms address this state estimation problem, but they require models of process dynamics and sensory observations and the respective noise characteristics of these models. Recently, multiple works have demonstrated that these models can be learned by end-to-end training through differentiable versions of recursive filtering algorithms. In this work, we investigate the advantages of differentiable filters (DFs) over both unstructured learning approaches and manually-tuned filtering algorithms, and provide practical guidance to researchers interested in applying such differentiable filters. For this, we implement DFs with four different underlying filtering algorithms and compare them in extensive experiments. Specifically, we (i) evaluate different implementation choices and training approaches, (ii) investigate how well complex models of uncertainty can be learned in DFs, (iii) evaluate the effect of end-to-end training through DFs and (iv) compare the DFs among each other and to unstructured LSTM models.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Combining Learning and Algorithmsmentioning

confidence: 99%

How to train your differentiable filter

2021

View full text Add to dashboard Cite

show abstract

“…The art of improving the performance of any deep-learning framework is a process of iterated refinements. Currently, there is no single ideal framework that addresses the discontinuous, impulsive and irregular patterns of behaviour associated with irregular-patterned complex sequential datasets [19,42]. These extreme datasets can be found in many different domains, including: health care, traffic, finance, such as stock prices, meteorology, such as rainfall data and so forth.…”

Section: Introductionmentioning

confidence: 99%

Complex Sequential Data Analysis: A Systematic Literature Review of Existing Algorithms

Dandajena

Venter

Ghaziasgar

et al. 2020

Conference of the South African Institute of Computer Scientists and Information Technologists 2020

View full text Add to dashboard Cite

This paper provides a review of past approaches to the use of deep-learning frameworks for the analysis of discrete irregularpatterned complex sequential datasets. A typical example of such a dataset is financial data where specific events trigger sudden irregular changes in the sequence of the data. Traditional deep-learning methods perform poorly or even fail when trying to analyse these datasets. The results of a systematic literature review reveal the dominance of frameworks based on recurrent neural networks. The performance of deep-learning frameworks was found to be evaluated mainly using mean absolute error and root mean square error accuracy metrics. Underlying challenges that were identified are: lack of performance robustness, non-transparency of the methodology, internal and external architectural design and configuration issues. These challenges provide an opportunity to improve the framework for complex irregular-patterned sequential datasets.

show abstract

“…In the non-convex environment, the RAD-A2C completed greater than 95% of 504 episodes over a range of obstructions and SNRs. There was very little gradient informa-505 tion available in the environments with more obstructions and thus the GS algorithm The PFGRU is an embedding of the BPF into a GRU architecture proposed by Ma 546 et al [26]. As in the BPF, there are a set of particles and weights used for filtering and…”

mentioning

confidence: 99%

“…The first component 561 is the mean squared loss between the mean particle and the predicted quantity. The 562 second component is the evidence lower bound (ELBO) loss that measures the difference in 563 distribution of the particle distribution relative to the observation likelihood, for more 564 details see [26]. The total loss is expressed as,…”

mentioning

confidence: 99%

Proximal Policy Optimization for Radiation Source Search

Proctor¹,

Teuscher²,

Hecht³

et al. 2021

Preprint

View full text Add to dashboard Cite

Rapid search and localization for nuclear sources can be an important aspect in preventing human harm from illicit material in dirty bombs or from contamination. In the case of a single mobile radiation detector, there are numerous challenges to overcome such as weak source intensity, multiple sources, background radiation, and the presence of obstructions, i.e., a non-convex environment. In this work, we investigate the sequential decision making capability of deep reinforcement learning in the nuclear source search context. A novel neural network architecture (RAD-A2C) based on the actor critic (A2C) framework and a particle filter gated recurrent unit for localization is proposed. Performance is studied in a randomized 20 x 20 m convex and non-convex environment across a range of signal-to-noise ratio (SNR)s for a single detector and single source. RAD-A2C performance is compared to both an information-driven controller that uses a bootstrap particle filter and to a gradient search (GS) algorithm. We find that the RAD-A2C has comparable performance to the information-driven controller across SNR in a convex environment and at lower computational complexity per action. The RAD-A2C far outperforms the GS algorithm in the non-convex environment with greater than 95% median completion rate for up to seven obstructions.

show abstract

Particle Filter Recurrent Neural Networks

Cited by 56 publications

References 14 publications

How to train your differentiable filter

How to train your differentiable filter

Complex Sequential Data Analysis: A Systematic Literature Review of Existing Algorithms

Proximal Policy Optimization for Radiation Source Search

Contact Info

Product

Resources

About