Incremental Slow Feature Analysis: Adaptive Low-Complexity Slow Feature Updating from High-Dimensional Input Streams

By Varun Raj Kompella, Matthew D. Luciw, J. Urgen Schmidhuber. 2012

60 citations recorded by the source corpus.

Subfields: Cognitive Neuroscience, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

1. Marco Mirolli
2. Gianluca Baldassarre
3. Vieri Giuliano Santucci
4. Kimberly L. Stachenfeld
5. David G. T. Barrett
6. David Pfau
7. Stig Petersen
8. Ashish Agarwal

In Signal Processing

1. Ashish Agarwal
2. Stig Petersen
3. David Pfau
4. David G. T. Barrett
5. Kimberly L. Stachenfeld
6. Dilip Ravi Kiran
7. Mathew Thomas
8. Ranjith Parakkal

Citations in the corpus

• Deep learning in neural networks: An overview (2014, methodology): “010; Riedmiller et al., 2012) where purely visual inputs were compactly encoded by deep autoencoders (Sec. 5.8, 5.15). RL combined with UL based on Slow Feature Analysis (Wiskott and Sejnowski, 2002; Kompella et al., 2012) enabled a real humanoid robot to learn skills from raw high-dimensional video”
• On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models (2015, methodology): “RL combined with unsupervised learning based on Slow Feature Analysis [318, 131] enabled a humanoid robot to learn skills from raw video streams [154].”
• Spectral Inference Networks: Unifying Deep and Spectral Learning (2018, methodology): “SFA has primarily been applied to train shallow or linear models, and when trained on deep models is typically trained in a layer-wise fashion, rather than end-to-end (Kompella et al., 2012; Sun et al., 2014).”
• GRAIL: A Goal-Discovering Robotic Architecture for Intrinsically-Motivated Learning (2016, background): “For example, the identification of interesting events could be provided by more sophisticated algorithms such as slow features analysis (SFA) [39], [94], that is able to identify complex events from a visual stream input (see [40] for an example where SFA and IMs are integrated to foster skill acqui”
• An overview of deep learning based methods for unsupervised and semi-supervised anomaly detection in videos (2018, methodology): “Authors in [66] propose an incremental application of SFA that updates slow features incrementally.”
• Construction of approximation spaces for reinforcement learning (2013, methodology): “(2010) hierarchical nonlinear slow feature analysis (SFA, Wiskott and Sejnowski, 2002) and Luciw and Schmidhuber (2012) incremental SFA (Kompella et al., 2012) to represent the underlying state space.”
• Spectral Inference Networks: Unifying Spectral Methods With Deep Learning (2018, methodology): “SFA has primarily been applied to train shallow models, and when trained on deep models is typically trained in a layer-wise fashion, rather than end-to-end [19, 32].”