Machine learning - a probabilistic perspective

By Kevin P. Murphy, Kevin P. Murphy. 2012

10,307 citations recorded by the source corpus.

Subfields: Probabilistic & Bayesian Methods

How researchers see this paper

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

In Probabilistic & Bayesian Methods

Citations in the corpus

Deep Generative Stochastic Networks Trainable by Backprop (2013, background): “a simple (typically factorial) approximate posterior q x(h) that is close to P(hjx), and usually involves an iterative optimization procedure. See a recent machine learning textbook for more details (Murphy, 2012). In addition, a challenge related to inference is sampling (not just from P(h jx) but ”
Neural Turing Machines (2014, background): “The optimal estimator for the problem can be determined by Bayesian analysis (Murphy, 2012): where c is the ﬁve bit previous context, B is the value of the next bit and N 0 and N 1 are respectively the number of zeros and ones observed after c so far in the sequence.”
Flow Contrastive Estimation of Energy-Based Models (2019, methodology): “The gradient of the second term is similar to reverse Kullback-Leibler divergence between q α and p θ , or variational approximation of p θ by q α , which forces q α to chase the modes of p θ [52, 12].”
Structured Convolutional Kernel Networks for Airline Crew Scheduling (2021, neutral): “Since the negative log-likelihood can be written as − log(p(yi|xi;w)) = Φi(−Ai w) (Murphy, 2012), the primal objective function to minimize over w ∈ R becomes P (w) := λ2 ‖w‖ 2 2+ 1 n ∑n i=1 Φi(−Ai w).”
Beyond CCA: Moment Matching for Multi-View Models (2016, neutral): “In FA and PPCA, one can enforce additional constraints to recover unique factor loading matrices (see, e.g., Murphy, 2012).”
Towards the Science of Security and Privacy in Machine Learning (2016, methodology): “Examples include the hinge loss [23] used in SVMs or the cross-entropy loss [3].”
Initialization and Regularization of Factorized Neural Layers (2021, methodology): “(2019) suggest compression-rate scaling (CRS), which scales weight-decay using the reduction in parameter count; this is justified via the usual Bayesian understanding of `2-regularization (Murphy, 2012).”
Meta-learning with differentiable closed-form solvers (2018, methodology): “When one applies Newton’s method to logistic regression, the resulting algorithm takes a familiar form — it consists of a series of weighted least squares (or ridge regression) problems, giving it the name Iteratively Reweighted Least Squares (IRLS) (Murphy, 2012, Chapter 8.3.4).”
Unified Contrastive Learning in Image-Text-Label Space (2022, methodology): “Connections to Cross-Entropy [46] We note the proposed LBiC in (1) is closely related to the standard cross-entropy loss used in supervised image classification.”
Planning chemical syntheses with deep neural networks and symbolic AI (2017, background): “[43] Expansion procedure The expansion policy network and the in-scope ﬁlter network are then combined to a pipeline (Figure 2 b)).”
Stochastic variational inference (2012, methodology): “Statistical machine learning research has addressed some of these challenges by developing the ﬁeld of probabilistic modeling, a ﬁeld that provides an elegant approach to developing new methods for analyzing data (Pearl, 1988; Jordan, 1999; Bishop, 2006; Koller and Friedman, 2009; Murphy, 2012).”
Kernel Mean Embedding of Distributions: A Review and Beyonds (2016, methodology): “To ease the reading, we suggest non-expert readers to also consult elementary machine learning textbooks such as Bishop (2006), Schölkopf and Smola (2002), Mohri et al. (2012), and Murphy (2012).”
Decentralized Bayesian Learning over Graphs (2019, neutral): “Minimization performed in Equation (1) is referred to as Variational Inference (VI) and the minimand is referred to as the variational free energy [23, 10, 5, 4].”
Privacy-Preserving Distributed Linear Regression on High-Dimensional Data (2017, background): “Interestingly, solving systems of linear equations is a basic building block of many other machine learning algorithms [54].”
The Limitations of Deep Learning in Adversarial Settings (2015, background): “Deep learning can be partitioned in two categories, depending on whether DNNs are trained in a supervised or unsupervised manner [29].”
Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples (2016, methodology): “Upon completion of their training on collections of known input-label pairs ( (cid:126)x,(cid:126)y ), classiﬁers f make label predictions f ( x ) on unseen inputs (cid:126)x [18].”
Recent Progress on Generative Adversarial Networks (GANs): A Survey (2019, background): “In the AI community, machine learning [1] exerts a huge impact on different aspects of our daily life.”