An Input Output HMM Architecture

By Yoshua Bengio, Paolo Frasconi. 1994

How researchers see this paper

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

Citations in the corpus

Composing graphical models with neural networks for structured representations and fast inference (2016, methodology): “[1] [2] [3] [4] Gaussian mixture model Linear dynamical system Hidden Markov model Switching LDS [8,9] [10] Canonical correlations analysis admixture / LDA / NMF [6] [2] [5] Mixture of Experts Driven LDS IO-HMM Factorial HMM [7]”
Learning Precise Timing with LSTM Recurrent Networks (2003, methodology): “Hybrid HMM-RNN approaches (Bengio and Frasconi, 1995) may be able to combine the virtues of both methodologies, but to our knowledge have never been applied to the problem of precise event timing as discussed here.”
Global training of document processing systems using graph transformer networks (1997, unknown): “Learning has been the focus of another line of work, called Input-Output HMM, which can model distributions over sequences conditioned on an input sequence [1].”
Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge (2011, methodology): “Due to the general trend to multimodal processing and availability of more complex algorithms to this aim such as asynchronous Hidden Markov Models (HMM), Dynamic Bayesian Networks (DBN), general Graphical Models or Multidimensional Dynamic Time Warp (DTW) and Meta-Classiﬁcation (Bengio and Frasconi”
Template-warping based speech driven head motion synthesis (2013, background): “The input-output extension to Markov models [12] allows for modiﬁcation of the observables based on factors other than states and other parts of the observation.”
Gradient-based learning applied to document recognition (1998, methodology): “A different approach to graph-based trainable systems, called input–output HMM, was proposed in [104] and [105].”
Learning Dynamic Bayesian Networks (1997, methodology): “Like state-space models, HMMs can be augmented to allow for input variables [7, 4, 36].”
Hierarchical Recurrent Neural Networks for Long-Term Dependencies (1995, methodology): “Our earlier analysis (Bengio and Frasconi, 1995a) shows that the difficulty in representing and learning to represent context (i .e., learning what Zt should represent) revolves around equation 4.”
An Introduction to Hidden Markov Models and Bayesian Networks (2001, unknown): “2, given the evidence e = {X = x,Z = z}, P (Y |X = x,Z = z) ∝ [∫ P (Y |W )P (W )dW ] P (Z = z,X = x|Y ) (4) ∝ P (Y )P (Z = z|X = x, Y )P (X = x) (5)”
Adaptive computation and machine learning (1998, background): “Another set of models described in [40, 58] and discussed in a later section are the IOHMMs (input-output HMMs) (see figure 9.”
LSTM can Solve Hard Long Time Lag Problems (1996, background): “also often used as benchmark problems for recurrent nets [2, 19, 14, 11].”
Algorithms for Sequential Decision Making (1996, methodology): “[14] Y. Bengio and P. Frasconi.”
Learning to perceive the world as articulated: an approach for hierarchical learning in sensory-motor systems (1998, background): “(4); the second term represents the damping term which suppresses abrupt changes in the gate opening; g and g are parameters.”
Pattern Recognition and Machine Learning (2006, unknown): “Another example is the input-output hidden Markov model (Bengio and Frasconi, 1995), in which we have a sequence of observed variables u1, .”
Autonomous agents modelling other agents: A comprehensive survey and open problems (2017, background): “Similarly, Gold [123] trains an Input-Output Hidden Markov Model [36] to predict a player’s goal in an action-adventure game.”
Hybrid Modeling, HMM/NN Architectures, and Protein Applications (1996, background): “We recently found out that related ideas have been proposed independently in Bengio and Frasconi (1995), but from a different viewpoint in terms of input/outyut HMMs.”
Bridging Long Time Lags by Weight Guessing and \Long Short Term Memory" (1996, background): “See, e.g., [2, 32, 22, 17, 16].”