Many-Layered Learning

By Paul E. Utgoff, David John Stracuzzi. 2002

157 citations recorded by the source corpus.

How researchers see this paper

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

In Artificial Intelligence

Citations in the corpus

A Distributional Perspective on Reinforcement Learning (2017, background): “A recurring theme in artiﬁcial intelligence is the idea of an agent learning from a multitude of predictions (Caruana 1997; Utgoff & Stracuzzi 2002; Sutton et al. 2011; Jaderberg et al. 2017).”
On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models (2015, methodology): “As the history grows, to find an M with small bitsM + bitsH , step 4 uses sequential network construction: it regularly changes M ’s size by adding or pruning units and connections [111, 112, 8, 168, 59, 107, 304, 176, 141, 92, 143, 204, 53, 296, 106, 31, 57, 185, 283].”
A Agent Environment StatesActions Rewards Critic B Agent Internal Environment Rewards Critic External Environment Sensations StatesDecisions Actions " Organism " Figure 1 (2004, unknown): “This process will naturally produce what Utgoff and Stracuzzi [30] called “many-layered” learning in which the agent learns what is easy to learn first, then uses this knowledge to learn harder things.”
Extracting and composing robust features with denoising autoencoders (2008, unknown): “Stracuzzi, 2002).”
Introduction to the Special Issue on Meta-Learning (2004, background): “The idea is somewhat related to layered learning (Stone & Veloso, 2000; Utgoff & Stracuzzi, 2003), except tasks need not build on top of each other but simply interact with each other.”
Intrinsically Motivated Reinforcement Learning: A Promising Framework for Developmental Robot Learning (2005, background): “Layered learning (Stone 2000; Utgo & Stracuzzi 2002 ) means that we can use lowerlevel learning elements to learn useful structures, discretizations, and behavior that can help make higherlevel learning feasible, and allows for the interaction of several learning elements to generate complex adapti”
Greedy Layer-Wise Training of Deep Networks (2006, background): “…machines: shallow boosting: generally shallow multi-layer neural networks: usually shallow, can be deep? decision trees: deep but local estimators (curse of dim.) parametric graphical models: human-labor intensive Non-parametric ones can theoretically approximate any continuous function.”
Deep learning for AI (2021, background): “Although the intuition that deeper neural networks could be more powerful pre-dated modern deep learning techniques, 82 it was a series of advances in both architecture and training procedures, 15,35,48 which ushered in the remarkable advances which are associated with the rise of deep learning.”
Evolving Unsupervised Deep Neural Networks for Learning Meaningful Representations (2017, background): “In fact, DRs are the outcomes which have gone through nonlinear transformations from input data more than once [15], and are inspired by the mammalian hierarchical visual pathway [16–18].”
Scaling learning algorithms towards AI (2007, background): “See [Utgoff and Stracuzzi, 2002] for a previous discussion of this question in the context of learning architectures.”
Knowledge Transfer in Deep convolutional Neural Nets (2007, background): “One tends to need far more nodes to represent a given function with a shallow architecture than with a deep architecture.(15,16) This would seem to imply that deep nets would be superior to shallow ones, because they need fewer parameters.”
On layer-wise representations in deep neural networks (2013, background): “See Utgoff and Stracuzzi (2002) and Bengio and LeCun (2007) for a discussion.”
Proto-value Functions: A Laplacian Framework for Learning Representation and Control in Markov Decision Processes (2007, background): “Amarel’s ideas motivated much subsequent research on representation discovery (Subramanian, 1989; Utgoff and Stracuzzi, 2002), and many methods for discovering global state space properties like “bottlenecks” and “symmetries” have been studied (McGovern, 2002; Ravindran and Barto, 2003; Mannor et al”
Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion (2010, background): “…neural network research that the composition of several levels of nonlinearity would be key to efﬁciently model complex relationships between variables and to achieve better generalization performance on difﬁcult recognition tasks (McClelland et al., 1986; Hinton, 1989; Utgoff and Stracuzzi, 2002).”
State-of-the-Art Deep Learning: Evolving Machine Intelligence Toward Tomorrow’s Intelligent Network Traffic Control Systems (2017, background): “Deep learning architectures are bio-inspired in the sense that brains have a deep architecture [22], [23], and computers can emulate such deep architectures.”