Machine Learning for Sequential Data: A Review

By Thomas G. Dietterich. SSPR/SPR, 2002

767 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/23c3953fb45536c9129e86ac7a23098bd9f1381d

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

In Information Systems

Citations in the corpus

Machine learning for Internet of Things data analysis: A survey (2017, background): “In these cases, data points consist of sequences of ð x ; y Þ pairs rather than being drawn in an i.i.d. manner from a joint distribution p ð x ; y Þ ; and such the sequences exhibit signi ﬁ cant sequential correlations [55,127].”
Online Learning: A Comprehensive Survey (2018, background): “Sequential learning is mainly concerned with learning from sequential training data (Dietterich, 2002), formulated as follows: a learner trains a model from a collection of N training data pairs {(x(i),y(i)), i = 1, . . . , N} where x(i) = (xi1, xi2, . . . , xiNi) is an Ni-dimensional instance…”
Exploration from Demonstration for Interactive Reinforcement Learning (2016, methodology): “We circumvent this problem by storing a batch data matrix, update it with new observations using a (FIFO) sliding window and compute the required parameters [12] online.”
Stacked Sequential Learning (2005, methodology): “Another plausible sequential learning method to apply to this task are maximum-entropy Markov models (MEMMs) [9], also called maximum-entropy taggers [11], conditional Markov models [7], and recurrent sliding windows [4].”
Empirical prediction models for adaptive resource provisioning in the cloud (2012, methodology): “In addition, we incorporate sliding window [15] and crossvalidation [11,12] techniques in the training and prediction stages and consequently evaluate the validity of the learning algorithms for prediction accuracy using statisticalmetrics, as discussed in the following sections.”
A review of unsupervised feature learning and deep learning for time-series modeling (2014, background): “Analysis of 5 time-series data has been the subject of active research for decades (Keogh 6 and Kasetty, 2002; Dietterich, 2002) and is considered by Yang and Wu 7 (2006) as one of the top 10 challenging problems in data mining due to 8 its unique properties.”
Conditional random fields for activity recognition (2007, background): “In the sequential supervised learning problem [23], our task is to construct a hypothesis (or classifier) h from a training set that contains a set of labeled sequences.”
A Linear Programming Formulation for Global Inference in Natural Language Tasks (2004, methodology): “Our approach allows us to efficiently incorporate domain and task specific constraints at decision time, resulting in significant improvements in the accuracy and the “human-like” quality of the inferences.”
Sequence classification for credit-card fraud detection (2018, methodology): “A general review on the properties of sequential classification tasks and suitable methods to cope with them can be found in [Dietterich, 2002].”
Max-Margin Markov Networks (2003, background): “However, it is well-known that this approach fails to exploit a lot of significant information [5].”
Twenty Years of Mixture of Experts (2012, background): “Such problems were discussed by Dietterich in [97].”
Policies for Active Learning from Demonstration (2017, background): “Therefore, LfD is related to sequential machine learning (Dietterich 2002).”
An introduction to domain adaptation and transfer learning (2018, background): “If the classifier receives feedback after each time step, the drift can be modeled and the next timestep’s shift can be predicted [55,72,57,41].”