
BOOKS
An Introduction to Statistical Learning with Applications in R (second edition)
by Gareth James, Daniela Witten, Trevor Hastie and Robert Tibshirani (August 2021)
3 new chapters (+179 pages), including Deep Learning
Book Homepage

Computer Age Statistical Inference:Algorithms, Evidence and Data Science
by Bradley Efron and Trevor Hastie (August 2016)
Book Homepage
pdf
(8.5 Mb, corrected online)

Statistical Learning with Sparsity: the Lasso and Generalizations
by Trevor Hastie, Robert Tibshirani and Martin Wainwright (May 2015)
Book Homepage
pdf (10.5Mb, corrected online)

The Elements of Statistical Learning:
Data Mining, Inference, and Prediction (Second Edition)
by Trevor Hastie, Robert Tibshirani and Jerome Friedman (2009)
Book Homepage
pdf (13Mb, correct. 12th print)



PAPERS
The research reported here was partially supported by grants from the
National Science Foundation and the National Institutes of Health.
For medical papers see also
2021
 Zijun Gao and Trevor Hastie.
LinCDE: Conditional density estimation via Lindsey's method.
Lindsey's method allows for smooth density estimation by turning the density problem into a Poisson GLM.
In particular, we represent an exponential tilt function in a basis of natural splines, and use discretization to deal with the normalization. In this paper we extend the method to conditional density estimation via trees and then gradient boosting with trees.

Yosuke Tanigawa, Junyang Qian, Guhan Venkataraman, Johanne Justesen, Ruilin Li, Robert Tibshirani, Trevor Hastie, Manuel Rivas.
Significant Sparse Polygenic Risk Scores across 428 traits in UK Biobank.
In this survey across the more than 1,600 traits in the UK Biobank, we report 428 strongly significant (p<2.5e5) sparse polygenic risk models, computed by the snpnet lasso model developed by this team.
 Swarnadip Ghosh, Trevor Hastie and Art Owen.
Scalable logistic
regression with crossed random effects.
We develop an approach for fitting crossed randomeffect
logisitic regression models at massive scales, with
applications in ecommerce. We adapt a procedure
of Schall (1991) and backfitting algorithms to achieve
O(n) algorithms.
 Zijun Gao and Trevor Hastie.
DINA: Estimating
Heterogenous Treatment Effects in Exponential Family and Cox
Models. We extend the Rlearner framework to exponential
families and the Cox model. Here we define the treatment
effect to be the difference in natural parameter or
DINA.

Stephen Bates, Trevor Hastie and Rob Tibshirani.
Crossvalidation:
what does it estimate and how well does it do it?
Although CV is ubiquitous in data science, some of its
properties are poorly understood. In this paper we argue
that CV is better at estimating expected prediction error rather than
the prediction
error for the particular model fit to the training set. We also provide a
method for computing the standard error of the CV estimate,
which is bigger than the commonly used naive estimate
which ignores the correlations in the folds.
 Elena Tuzhilina, Leonardo Tozzi and Trevor Hastie.
Canonical Correlation Analysis in high dimensions with structured regularization.We develop structurally regularized versions of CCA for very highdimensional MRI images from neuroscience experiments. To appear, Statistical Modelling, 2021.

J. Kenneth Tay, Balasubramanian Narasimhan and Trevor Hastie.
Elastic Net
Regularization Paths for All Generalized Linear Models.
This paper describes some of the substantial enhancements to
the glmnet R package ver 4.1+. All
programmed GLM families are accommodated through a
family() argument. We also discuss relaxed fits,
and facilities for modeling stop/start data and strata in
survival models.
2020
 Benjamin HaibeKains et
al.
Transparency and reproducibility in artificial
intelligence. A note in Matters arising, Nature,
where we are critical of the lack of transparency in a
highprofile paper on using AI for breastcancer screening.
 Lukasz Kidzinski, Francis Hui, David Warton and Trevor
Hastie.
Generalized Matrix Factorization We fit largescale
generalized linear latent variable models (GLLVM) by using
efficient methodology adapted from matrix
factorization. Designed with large scale ecology studies in
mind with tens of thousands of species and/or locations, these
methods provide good approximations to the stateoftheart
variational Bayes methods that do not scale gracefully to such
sizes.
 Swarnadip Ghosh, Trevor Hastie and Art
Owen.
Backfitting for large scale crossed random effects
regressions
Regression models with crossed randomeffects
can be very expensive to compute in large scale
applications (eg Ecommerce with millions of customers and
many thousands of products). We adapt a backfitting
algorithm to problems of this kind and achieve convergence (provably)
linear in the number of observations. to appear, Annals of Statistics 2021

Elena Tuzhilina, Trevor Hastie and Mark Segal.
Principal curve approaches for inferring 3D chromatin architecture
We adapt metric scaling algorithms to approximate the 3D
conformation of a chromosome based on contact maps. Biostatistics, Nov 2020.
 Trevor Hastie.
Ridge
Regularization: an Essential Concept in Data Science
This paper is written by request to celebrate the 50th
anniversary of the first ridgeregression paper by Hoerl
and Kennard (1070), Technometrics. In it I gather together
some of my favorite aspects of ridge, and also touch on
the "doubledescent"
phenomenon. Technometrics
online link
 Kenneth Tay, Nima Aghaeepour, Trevor Hastie, Robert
Tibshirani.
Featureweighted
elastic net: using "features of features" for better
prediction We often have metadata that informs us about
features to be used for supervised learning. For example,
genomic features may belong to different pathways. Here we develop a
method for exploiting this information to improve prediction. Statistica Sinica 2021.
 Junyang Qian, Yosuke Tanigawa, Ruilin Li, Robert
Tibshirani, Manuel A Rivas, Trevor
Hastie. LargeScale
Sparse Regression for Multiple Responses with Applications to UK
Biobank
We develop methodology to build large sparse polygenic risk score
models using multiple phenotypes (multitask learning).
Our methodology builds on our early snpnet lasso models
for single phenotypes, and includes missing phenotype imputation
and adjustment for confounders.
Here is a link to the code and
scripts used in the paper
https://github.com/junyangq/scripts_multiSnpnet_paper
2019
 Trevor Hastie, Andrea Montanari, Saharon Rosset and Ryan
Tibshirani.
Surprises in HighDimensional Ridgeless Least Squares
Interpolation.
Interpolating fitting algorithms have
attracted growing attention in machine learning, mainly because
stateofthe art neural networks appear to be models of this
type. In this paper, we study minimum L2norm ("ridgeless")
interpolation in highdimensional least squares regression. We
consider both a linear
model and a version of a neural network. We recover several
phenomena that have been observed in largescale neural networks and
kernel machines, including the "double descent" behavior of the
prediction risk, and the potential benefits of overparametrization.
 Zijun Gao, Trevor Hastie and Rob Tibshirani.
Assessment
of heterogeneous treatment effect estimation accuracy via
matching
We address the difficult problem of assessing the performance
of an HTE estimator. Our approach has several novelties: a
flexible matching metric based on randomforest proximity
scores, an optimized matching algorithm, and a match then
split crossvalidation scheme.
Statistics in Medicine, April 2021
2018
 Junyang Qian, Yosuke Tanigawa, Wenfei Du, Matthew Aguirre,
Chris Chang, Robert Tibshirani, Manuel A. Rivas, Trevor Hastie.
A Fast and Scalable Framework for Largescale and
Ultrahighdimensional Sparse Regression with Application to the UK
Biobank.
PLOS Genetics October 2020.
We develop a scalable lasso
algorithm for fitting polygenic risk scores at GWAS scale. There is also a BiorXiv version. Our R
package snpnet combines efficient batchwise strongrule
screening with glmnet to fit lasso regularization paths on
phenotypes in the UK Biobank data.
Here is a link to the code and
scripts used in the paper
https://github.com/junyangq/scripts_snpnet_paper
 Lukasz Kidzinski and Trevor Hastie
Longitudinal data
analysis using matrix completion
We use a regularized form of matrix completion to fit functional
principal component models, and extend these to other multivariate
longitudinal regression models. We have an R package
fcomplete
which includes three vignettes demonstrating how it can be used.
2017
 Trevor Hastie, Rob Tibshirani and Ryan
Tibshirani
Extended Comparisons of Best Subset Selection, Forward Stepwise
Selection, and the Lasso
This paper is a followup to "Best Subset Selection from a Modern Optimization Lens"
by Bertsimas, King, and Mazumder (AoS, 2016). We compare these
methods using a broad set of simulations that cover typical
statistical applications. Our conclusions are that bestsubset
selection is mainly needed in very high signaltonoise regimes, and the relaxed lasso is the overall winner.
 Scott Powers, Junyang Qian, Kenneth Jung, Alejandro Schuler,
Nigam Shah, Trevor Hastie and Robert
Tibshirani. Some methods for
heterogeneous treatment effect estimation in highdimensions
We develop some new methods for estimating personalized treatment
effects from observational data: causal boosting and causal MARS.
Statistics
in Medicine. (January 2018).
2016
 Scott Powers, Trevor Hastie and Rob Tibshirani. Nuclear penalized multinomial regression
with an application to predicting atbat
outcomes in baseball. Here we use a convex formulation of the
reducedrank multinomial model, in a novel application using a
large dataset of baseball statistics. In special
edition "Statistical Modelling for Sports Analytics", Statistical Modelling, vol. 18, 56: pp. 388410.
 Qingyuan Zhao and Trevor Hastie
Causal Interpretations of BlackBox Models.
We draw connections between Friedman's partial dependence plot and
Pearl's backdoor adjustment to explore the possibility of
extracting causality statements after fitting complex models
by machine learning algorithms. Finally published 2019 ASA/JBES, 39(1) 2019
 Nicholas Boyd, Trevor Hastie, Stephen Boyd, Benjamin Recht,
Michael Jordan.
Saturating Splines and Feature Selection We use a convex
framework based on TV (total variation) penalty norms for
nonparametric regression. Saturation for degreetwo splines
requires the solution extrapolates as a constant beyond the range
of the data. This along with an additive model formulation leads
to a convex path algorithm for variable selection and smoothing
with generalized additive
models. JMLR 18(197):132, 2018
2015
 Scott Powers, Trevor Hastie and Robert Tibshirani Customized training with an application to mass spectrometric imaging of cancer tissue. Annals of Applied Statistics 9(4) (2015), 17091725.
 Rakesh Achanta and Trevor Hastie Telugu OCR Framework using Deep Learning.
We build an endtoend OCR system for Telugu script, that segments the text image, classifies the characters and extracts lines using a language model.The classification module, which is the most challenging task of the three, is a deep convolutional neural network.
 Jingshu Wang, Qingyuan Zhao, Trevor Hastie and Art
Owen. Confounder
Adjustment in Multiple Hypotheses Testing. (accepted,
Annals of Statistics, 2016).
We present a unified framework for analysing different proposals for adjusting for confounders in multiple testing (e.g. in genomics). We also provide an R package
cate on CRAN that implements these different approaches. The vignette shows some examples of how to use it.
 Alexandra Chouldechova and Trevor Hastie
Generalized Additive Model Selection A method for selecting terms in an additive model, with sticky selection between null, linear and nonlinear terms, as well as the amount of nonlinearity. The R package gamsel has been uploaded to CRAN.
2014
 Trevor Hastie, Rahul Mazumder, Jason Lee and Reza Zadeh.
Matrix Completion and LowRank SVD via Fast Alternating Least Squares We develop a new method for matrix completion, that improves upon our earlier softImpute algorithm, as well as the popular ALS algorithm.
JMLR 2015 16 33673402. We have also incorporated this method in our
R package softImpute.
 Ya Le, Trevor
Hastie. Sparse Quadratic
Discriminant Analysis and Community Bayes We provide a
framework for generalizing naive Bayes classification using sparse
graphical models.
Naive Bayes assumes conditional on the response, the variables are
independent. We relax that, and assume their conditional dependence
graph is shared and sparse.
 William Fithian, Jane Elith, Trevor Hastie, David
A. Keith. Bias Correction
in Species Distribution Models: Pooling Survey and Collection Data
for Multiple Species We develop methods for teasing out
observerbias in multispecies presenceonly data. (submitted,
arXiv:1403.7274) ( Methods for Ecology and Evolution,
October 10, 2014). Github link to the R package multispeciesPP
 Will Fithian and Trevor Hastie.
Local CaseControl Sampling: Efficient Subsampling in Imbalanced Data Sets (arXiv). Modern classification tasks are often extremely unbalanced  a situation where casecontrol sampling can be very helpful. This paper discusses the bias of CC sampling, and proposes a simple twostage subsampling procedure for removing this bias, and in cases dramatically improving the efficiency.
Annals of Statistics 2014, Vol. 42, No. 5, 16931724.
projecteuclid.org/euclid.aos/1410440622
2013
 Lucas Janson, Will Fithian, and Trevor Hastie. Effective Degrees of Freedom: a Flawed Metaphor. The popular covariance formula for df gives some surprising results, like df>p in forward stepwise.
May 2015, Biometrika
 Stefan Wager, Trevor Hastie and Bradley Efron. Confidence Intervals for Random Forests: the Jackknife and the Infinitessimal Jackknife. We use ideas related to OOB errors to compute standard errors for bagging and random forests. Two approaches are presented, one based on the jacknife, the other on the infinitessimal jacknife. We study the bias of these estimates, as well as montecarlo errors. .
(JMLR 2014, 15 16251651)
An R package is available for computing these estimates, currently residing on Stefan Wager's GitHub space; see the example.R file.
 David Warton, Bill Shipley and Trevor Hastie. CATS regression  a modelbased approach to studying traitbased community assembly. We show how to use GLMs to fit community models, which are traditionally fit by maximum entropy. Apart from being a convenient platform for model fitting, a all the usual summaries, statistics and extensions of GLMs are available. ( Methods of Ecology and Evolution, September 29, 2014)
 Hristo Paskov, Robert West, John Mitchell and Trevor Hastie. Compressive Feature Learning We use an unsupervised convex ducument compression algorithm to derive a sparse kgram representation for a corpus of documents. This same dictionary, in the spirit of "deep learning", is as good as the original kgram representation for document classification tasks. To appear, NIPS 2013.
 My first blog post with Will Fithian. This post refers to concerns that were raised about crossvalidation.
 Noah Simon, Jerome Friedman and Trevor Hastie. A Blockwise Descent Algorithm for Grouppenalized Multiresponse and Multinomial Regression. We use the group lasso in the context of multinomial and multiresponse regression. Each variable has multiple coefficients for the different responses, and they each get selected via a group lasso penalty. Our code is an efficient implementation of block coordinate descent, and is built into the glmnet package. (submitted, on arXiv).
 Michael Lim and Trevor Hastie. Learning interactions via hierarchical grouplasso regularization We use the overlap group lasso in the context of a linear model to test for interactions. Our methodology can handle qualitative as well as quantitative variables. Our R package
glinternet can fit linear and logistic regression models. Optimized code can handle thousands of variables (our largest example had > 20K 3level factors). arXiv:1308.2719 ( JCGS 2014, online access)
 Michael Jordan et al. Frontiers in Massive Data Analysis.
This 129 page document is the report produced by the Committee on the Analysis of Massive Data. This committee was established by the National Research Council of the National Acadamies, and met 4 times over 20112012 in Washington and California. Michael Jordan was the chair of the 18 member committee, made up of statisticians (5), computer scientists and mathematicians. I was a member of the committee, and was jointly responsible for Chapter 7 with David Madigan, although all committee members provided input to all chapters as well.
 Trevor Hastie and Will Fithian.
Inference from Presenceonly Data: the Ongoing Controversy. This short paper argues strongly against the use of rigid parametric logistic regression models to make inferences from presenceonly data in Ecology. Essentially the rigidity of the model manufactures information that is not present in the data. Ecography (2013, editors choice)
Video interview with David Warton at Ecostats conference at UNSW in Sydney in July 2013. David was wearing his editor's hat for
Methods in Ecology and Evolution, and the discussion centered on this paper.
 Noah Simon, Jerome Friedman, Trevor Hastie and Rob Tibshirani.
The Sparse Group Lasso
By mixing L1 penalties with grouplasso L2 penalties, we achieve a sparse group lasso where some members of a group can end up being zero.
JCGS, May 2013, 22(2), pages 231245.
 Jianqiang Wang and Trevor Hastie.
Boosted VaryingCoefficient Regression Models for Product Demand Prediction.
We use the varying coefficient paradigm to fit a market segmented product demand model, with boosted regression trees as the nonparametric component. JCGS (online access) March 2013.

Julia Viladomat, Rahul Mazumder, Alex McInturff, Douglas McCauley and Trevor Hastie
Assessing the significance of global and local
correlations under spatial autocorrelation; a nonparametric approach.
Variables collected over a spatial domain often exhibit strong spatial autocorrelation.
When such variables are used in a regression, pairwise correlation analysis, or in the popular geographically weighted regression, it can be difficult to assess significance. We propose a general approach based on randomization followed by smoothing to restore the spatial correlation structure.
R code used in the paper. (Biometrics, Jan 2014)
2012
 Will Fithian and Trevor Hastie. Finitesample equivalence in statistical models for presenceonly data. We show that a lot of different approaches to presenceonly data are the same, in particular inhomogenous poisson proccesses, maxent, and naive logistic regression (when weighted appropriately).
(AoAS 2013 7 (4) 19171939 ).
 Jason Lee and Trevor Hastie. Learning Mixed Graphical Models.
We use grouplasso regularized pseudo likelihood for learning the structure of a graphical model with mixed discrete and continuous variables. Our model respects the symmetry imposed by a Markov random field representation  each of the potentials gets a vote from a pair of regression models (Gaussian, logistic or multinomial), where each of the pair of variables is the response and predictor. (in Arxiv), JCGS 24(1) 2015.
Go to Jason Lee's webpage for matlab code and a demo.

Rahul Mazumder, Jerome Friedman and Trevor Hastie. Sparsenet R package on CRAN. Fits sparse solution paths for linear models (squareerror loss) using coordinate descent with MC+ penalty family.
Software is very fast, and can handle many thousands of variables. Functions for crossvalidation, prediction, plotting etc. Based on algorithms described in SparseNet
:
Coordinate Descent with NonConvex Penalties. JASA 2011, 106(495) 11251138.
2011
 Rahul Mazumder and Trevor Hastie
Exact Covariance Thresholding into Connected
Components for largescale Graphical Lasso A block screening rule for graphical models that can dramatically speed up computations, and allow for distributed computing of models on a much larger scale. Original copy arXiv 8/18/2011 published JMLR March 2012 13, 723736.
 Rahul Mazumder and Trevor Hastie
The Graphical Lasso: New Insights and Alternatives (arXiv 11/23/2011, published November 2012) We examine the glasso algorithm for solving the graphical lasso problem. We show that it solves the dual problem,
where the optimization variable is the covariance rather than the precision matrix. We propose a similar primal algorithm, which appears to be superior in speed and other propertiers. Electronic Journal of Statistics 20126 21252149.
R package dpglasso
 Gen Nowak; Trevor Hastie; Jonathan R. Pollack; Robert Tibshirani
A fused lasso latent feature model for analyzing multisample aCGH data.
Biostatistics (advanced access June 2011). A new approach for modeling multisample CGH data, that exploits the similarities in copynumber variation at the same locations in the genome. Here is the
online link at the Biostatistics journal website.
The R package FLLat is available from CRAN.
 Noah Simon, Jerome Friedman, Trevor Hastie and Rob Tibshirani
Regularization Paths for Cox's Proportional Hazards Model via Coordinate Descent.
We develop the tools needed to include the Cox proprtional hazard's model in GLMNET.
Journal of Statistical Software 39(5), 113 (2011).
2010
 Rob Tibshirani, Jacob Bien, Jerome Friedman, Trevor Hastie, Noah Simon, Jonathan Taylor and Ryan Tibshirani:
Strong
rules for discarding predictors in lassotype problems. We develop
rules for screening predictors for lasso and elasticnet penalized
models. When p>>n, this can result in large computational savings,
without any loss in accuracy. JRSS B (2012) 74

Jane Elith, Steven Phillips, Trevor Hastie, Miroslav Dudik, Yung En
Chee and Colin Yates A
statistical explanation of Maxent for Ecologists
Maxent is a method for modeling species prevalence with
presencebackground data. This paper explains maxent using the language of a
statistical models. Diversity and Distributions November 2010

Jerome Friedman, Trevor Hastie and Rob Tibshirani:
Applications of the lasso and grouped lasso to the estimation of sparse graphical models
We develop efficient algorithms for fitting sparse, undirected graphical models. These are specially designed for the "large p" situation.

Jerome Friedman, Trevor Hastie and Robert Tibshirani:
A Note on the Group Lasso and a Sparse Group Lasso.
We develop a group lasso with both sparsity of groups and sparsity within groups.
We also develop coordinatewise algorithms for fitting the both cases.
2009
 Rahul Mazumder, Jerome Friedman and Trevor Hastie:
SparseNet
:
Coordinate Descent with NonConvex Penalties. JASA 2011, 106(495) 11251138. Nonconvex penalties
produce sparser models than the LASSO, but pose difficulties for
optimization. We propose a structured algorithm using coordinate
descent which finds good solutions with guaranteed convergence
properties.
Appendix
for SparseNet paper with extra figures and some additional technical proofs.
sparsenet R package available from CRAN (Feb 2012)
 Rahul Mazumder, Trevor Hastie and Rob Tibshirani:
Spectral Regularization Algorithms for Learning Large
Incomplete Matrices. We develop
an iterative algorithm for matrix completion using nuclearnorm regularization. JMLR 2010 11 22872322
MATLAB package SoftImpute for matrix completion (zip archive). R package to appear soon.

Daniela Witten, Rob Tibshirani and Trevor Hastie:
A penalized matrix
decomposition, with applications to sparse canonical correlation analysis and
principal components Biostatistics 10(3)

Trevor Hastie, Robert Tibshirani and Jerome Friedman, Elements
of Statistical Learning: Data Mining, Inference and
Prediction (Second Edition). February, 2009. 745 pages in full color. SpringerVerlag, New York.
This second edition adds 4 new chapters: Random Forests, Ensemble Learning, Undirected Graphical Models, and High Dimensional Problems: p>>N. For more details see
ESL book homepage.
In an agreement with Springer, we are able to offer for free the
ESL book pdf (8.2M).
2008
 Tong Tong Wu, Yi Fang Chen, Trevor Hastie, Eric Sobel and Kenneth
Lange:
Genomewide
Association Analysis by Lasso Penalized Logistic
Regression
We develop efficient computational procedures for screening
largescale genomewide association studies.
Bioinformatics 25(6): 714721, 2009.

Ping Li, Ken Church and Trevor Hastie:
One
sketch for all: theory and application of conditional
random sampling Nips08 proceedings.

Jerome Friedman, Trevor Hastie and Robert Tibshirani:
Regularization Paths for Generalized Linear Models via Coordinate Descent.
We use coordinate descent to develop regularization paths for linear,
logistic and multinomial regression models. Our algorithms use the
"elastic net" penalties of Zou and Hastie (2005), and create the
path for a grid of values of the penalty parameter lambda.
Journal of Statistical Software, 33(1), 2010
The R
package glmnet is available from
CRAN
A matlab wrapper for the glmnet fortran code, written by Hui Jiang.

Jane Elith, John Leathwick and Trevor Hastie:
A
working guide to boosted regression trees (2008) Journal of
Animal Ecology, 77 802813. Here are the
online supplement materials, along with the associated
zip
file.
In 2004 received award for most highly cited paper in any of the British Ecological Society journals in the past 5 years.

Line Clemmensen, Trevor Hastie, Daniela Witten and Bjarne Ersboll:
Sparse Discriminant Analysis.
We extend penalized linear and mixture discriminant analysis by
incorporating a lasso penalty to encourage sparseness. Technometrics (2011)
53(4) 406413.

John Leathwick, Jane Elith, W. Chadderton, D. Rowe and Trevor Hastie:
Dispersal, disturbance and the contrasting biogeophraphies of New
Zealand's diadromous and nondiadronous fish species.
An application of boosted regression trees in ecological mapping.
J. Biogeography 2008, 35 14811497.

Debashis Paul, Eric Bair, Trevor Hastie and Robert Tibshirani:
"Preconditioning" for feature selection and regression in
highdimensional problems
We show that supervised principal components followed by a variable selection
procedure is an effective approach for variable selection in very high dimension.
Annals of Statistics 36(4), 2008, 15951618.
2007

Jerome Friedman, Trevor Hastie and Robert Tibshirani,
Sparse inverse covariance estimation with the lasso.
We develop an effecient algorithm for solving the L1penalized
likelihood approach to sparse covariance estimation.
 Trevor Hastie
Comment on a paper in Statistical Science by Peter Bühlmann and Torsten Hothorn: Boosting
Algorithms:Regularization, Prediction and Model Fitting (2007) 22(4),477522.
 Jerome Friedman, Trevor Hastie and Robert Tibshirani,
Discussion of "Evidence contrary to the statistical view of boosting
(David Mease and Aaron Wyner)" Wyner and Mease show through
examples some counterintuitive results with boosting, that appear to
contradict our 2000 paper. We discount these claims by reversing their
results using shrinkage along with boosting. JMLR9 (2008) 5964.
 Jerome Friedman, Trevor Hastie, Holger Hoefling and Robert Tibshirani,
Pathwise Coordinate Optimization. We show how coordinate descent
algorithms can efficiently solve a number of popular regularized
optimization problems, creating an entire path of solutions. We
generalize this approach to derive an efficient algorithm for the
fused lasso, both one and twodimensional. Annals of Applied
Statistics (2007), 1(2), 302332.
 Ping Li, Trevor Hastie and Kenneth Church.
Nonlinear Estimators and Tail Bounds for Dimension Reduction in L1
using Cauchy Random Projections. We provide improved methods for approximating L1 distances in very
high dimensions, based on maximumlikelihood estimation in the Cauchy
family. JLMR 8, pp 24972532
 Ping Li and Trevor Hastie.
A Unified NearOptimal Estimator for Dimension Reduction in L_a
(0< a <= 2) Using Stable Random Projections. NIPS2007 poster
presentation
 Brad Efron, Trevor Hastie and Rob Tibshirani,
Discussion of the "Dantzig Selector" by Emmanuel Candes and Terrence
Tao.
Candes and Tao propose an alternative but similar procedure to the
lasso. This discussion appears alongside the orginal article in the
Annals of Statistics 35(6),
pp23582364.
 Trevor Hastie, Jonathan Taylor, Robert Tibshirani and Guenther
Walther,
Forward Stagewise Regression and the Monotone Lasso
We characterize the incremental forward stagewise procedure as a
monotone version of the lasso. Electronic Journal of Statistics
1 (2007).
2006
 Trevor Hastie and Ji Zhu,
Discussion of "Support Vector Machines with Applications" by
Javier M. Moguerza and Alberto Munoz, Statistical Science 21(3)
352357.
 Gill Ward, Trevor Hastie, Simon Barry, Jane Elith and John
Leathwick,
Presenceonly data and the EM algorithm. We develop a method for
fitting the twoclass logistic regression model using labeled data from
one class, a sample of unlabeled data, and knowledge of the
class prevalences.Biometrics,65(2)554563, 2009.
Download the beta version of Gill Ward's R package ecogbm 
ecogbm_1.01.tar.gz  for fitting boosted regression models for presenceonly data. See the
student section of my homepage for Gill Ward's thesis.
A presentation by Gill Ward based on this work "Making the Best Use of
Available Data: The PresenceOnly Problem in Ecology" won an
honorable mention award at the 2007 Joint Statistical Meetings.
 John Leathwick, Jane Elith and Trevor Hastie, Comparative performance of generalized
additive models and multivariate adaptive regression splines for
statistical modelling of species distributions. (2006) Ecological
Modelling 199 188196. This is a special issue of the
journal devoted to the workshop on
Advances in Predictive Species
Distribution Models held in Riederalp,Switzerland, 2004.
 Ping Li, Trevor Hastie and Kenneth Church,
Very
Sparse Random Projections. A method for approximating pairwise distances in
very highdimensional spaces. Best student paper, KDD06,
Philadelphia
 MeeYoung Park and Trevor Hastie,
Regularization Path Algorithms for Detecting Gene Interactions.
We develop a path algorithm for fitting the "cosso" models of Yuan &
Lin (2006) with logistic regression. This allows factors and interactions
to enter the model in a smooth way.
 MeeYoung Park and Trevor Hastie,
Penalized Logistic Regression for Detecting Gene Interactions.
A modified version of forwardstepwise logistic regression suitable
for screening large numbers of genegene interactions.
stepPlr:
R package for fitting PLR models.
 MeeYoung Park, Trevor Hastie and Rob Tibshirani, Averaged gene expressions for regression
A regression method that combines the lasso with hierarchical
clustering, intended for selecting groups of genes in microarray
problems.
Biostatistics (2007, 8, 212227).
R Software
for fitting these models.

Yaqian Guo, Trevor Hastie and Robert Tibshirani
Regularized
Discriminant Analysis and its Application in Microarrays.
A method, similar to shrunken centroids, for classification and
discrimination of microarrays, using regularized discriminant analysis
with gene selection. Biostatistics (in press; epub)
 MeeYoung Park and Trevor Hastie, An L1 Regularizationpath Algorithm for
Generalized Linear Models.
A generalization of the LARS algorithm for GLMs and the Cox
proportional hazard model. Since the coefficient
paths are piecewisenonlinear, approximations are made using the
predictorcorrector algorithm of convext optimization.
glmpath: R software package for fitting L1 regularized GLMs and
Cox models. (JRSSB 2007 (69, part 4), pages 659677 )

Ping Li, Trevor Hastie and Kenneth Church,
Improving Random Projections Using Marginal Information.
Methods for speeding up document search and characterization. Accepted
at Colt 2006
This paper draws on results in the following two technical
reports:

Rob Tibshirani and Trevor Hastie, Margin
Trees for Highdimensional Classification.
A treestructured representation for a multiclass SVM classifier.
 Hui Zou, Ji Zhu and Trevor Hastie,
New Multicategory Boosting Algorithms
Based on Multicategory FisherConsistent
Losses.
We provide some general requirements for multiclass marginbased
classifiers. Annals of Applied Statistics 2(4) pp 12901306, 2008).
2005
 Ji Zhu, Hui Zhou, Saharon Rosset and Trevor Hastie,
Multiclass Adaboost.
A multiclass generalization of the Adaboost algorithm, based on a
generalization of the exponential loss.
Finally
published in 2009 in
Statistics and Its Interface Volume 2 (2009) 349360.
Code on Ji Zhu's website
 J. Leathwick, J. Elith, M. Francis, T. Hastie, P. Taylor.
Variation in demersal fish species richness in the oceans surrounding
New Zealand: an analysis using boosted regression trees.
(Marine Ecology Progress Series, published in 2006).
A detailed analysis of species abundance using Poisson regression
with boosted regression trees. All analysis done using the gbm
package in R (Greg Ridgeway).
 J. Leathwick, D. Rowe, J. Richardson, J. Elith and T. Hastie,
Using multivariate adaptive regression splines to predict
the distributions of New Zealand's freshwater
diadromous fish.
Freshwater Biology 50 20342051.
Presenceabsence species data are modelled using a MARS along with GLM
in R.
 MeeYoung Park and Trevor Hastie,
Hierarchical Classification using Shrunken Centroids.
A technique for classification when the number of classes is large. It
produces an hierarchically structured classification rule, with the
hardesttoseparate classes at the terminal nodes.
2004

Hui Zou, Trevor Hastie, and Rob Tibshirani,
On
the "Degrees of Freedom" of the Lasso.
A technical paper that establishes that the number of nonzero
coefficients in a lasso model is unbiassed for the effective degrees
of freedom. Published in Annals of Statistics (2007),
35, 5, 21732192.

Philip Beineke, Trevor Hastie and Shivakumar Vaithyanathan,
The
Sentimental Factor: Improving Review Classification via HumanProvided
Information Proceedings ACL 2004, Barcelona. (ACL: Association of
Computational Linguistics)

Eric Bair, Trevor Hastie, Debashis Paul, and Robert Tibshirani
Prediction by Supervised Principal Components Published in
JASA 2006
101 No 473, pp 11137.

NIPS2004  The following papers were accepted for NIPS 2004:

Hui Zou, Trevor Hastie, and Rob Tibshirani.
Sparse
Principal Component Analysis. We present a new approach to
principal component analysis, that allows us to use an L1 penalty to
ensure sparseness of the loadings. Published in JCGS 2006 15(2):
262286. Software is available in R package elasticnet
available from CRAN.

Trevor Hastie, Saharon Rosset, Rob Tibshirani and Ji Zhu.
The
Entire Regularization Path for the Support Vector Machine.
JMLR, 5(Oct) 13911415.
An algorithm for computing the twoclass SVM solution for all possible
values of the regularization parameter C, at essentially the
computational cost
of a single SVM fit. Not only does this allow for efficient model
selection, but it also exposes the role of regularization for SVMs.
Several
MPEG movies show the sequence of solutions for different examples.
SvmPath
software package for R.

Trevor Hastie and Robert Tibshirani.
Efficient Quadratic Regularization for Expression
Arrays. Biostatistics (2004), 5(3), pp 329340. Computational tricks for a large class of linear
models fit by quadratic regularization.
2003

Hui Zou and Trevor Hastie.
Regularization
and Variable Selection via the Elastic Net (pdf). JRSSB (2005)
67(2) 301320. A compromise between ridge regression and the lasso,
with the computational advantages of the lasso. The elastic net
selects variables in correlated sets. An R package elasticnet
is available from CRAN.
See interview
on Essential Science Indicators.
Published minor
correction.

Jerome Friedman, Trevor
Hastie, Saharon Rosset, Rob Tibshirani and Ji Zhu.
Discussion
of three Boosting papers Annals of
Statistics, 2004, vol 32 (1) pp 102107. The three papers are by
(1) Wenxin Jiang, (2) Gabor Lugosi and Nicolas Vayatis, and (3)
Tong Zhang.

Saharon Rosset, Ji Zhu and Trevor Hastie.
Margin Maximizing Loss Functions
(accepted poster Nips 2003)

Ji Zhu, Saharon Rosset, Trevor Hastie and Rob Tibshirani.
1Norm Support Vector Machines
(accepted spotlight poster Nips 2003)
 Mu Zhu, Trevor Hastie and Guenther Walther.
Constrained Ordination Analysis with Flexible Response
Functions Constrained ordination via nonparametric
discriminant analysis. Ecological Modelling (2005),
187(4), 524536.

Francesca Dominici, Aidan McDermott and Trevor Hastie.
Improved Semiparametric Time Series Models of Air Pollution
and Mortality [pdf Technical Report] JASA, December 2004, 99(468), 938948.
[Sofware]
 Ji Zhu and Trevor Hastie
Classification of Gene Microarrays by Penalized Logistic
Regression. Biostatistics 5(3):427443.
R
Code (from Ji Zhu's website)
2002

Trevor Hastie and Robert Tibshirani.
Independent Component Analysis through
Product Density Estimation (ps file). A direct statistical approach to
ICA, using an attractive spline representation to model each of
the marginal densities.A more recent (Nov 2002)
talk (pdf)

Saharon Rosset, Ji Zhu and Trevor Hastie.
Boosting as a Regularized Path to a
Maximum Margin Classifier (pdf file) JMLR 5 (Aug 2004): 941973, 2004. We show that a
version of boosting fits a model by optimizing a
L1penalized loss function. This in turn shows that the
corresponding versions of Adaboost and Logitboost converge
to an "L1" optimal separating hyperplane.
 Support Vector Machines, Kernel Logistic
Regression and Boosting . Slides for talk given at Spring
research conference in Michigan, MCS2002 in Sardinia,
NPCONF2002 in Crete, and ASA2002 in New York.

Hongjuan Zhao, Trevor Hastie, Dr Michael L Whitfield, Prof AnneLise
BorresenDale and Dr. Stefanie S Jeffrey
Optimization
and evaluation of T7 based RNA linear amplification
protocols for cDNA microarray analysis BMC Genomics
2002, 3:31 (30 Oct 2002)
Biomed central
online

Robert Tibshirani, Trevor Hastie, Balasubramanian Narasimhan, and Gilbert Chu.
Class prediction by
nearest shrunken centroids, with applications
to DNA microarrays
(ps file)
(pdf file)
This is a more statistical version of the PNAS paper below.

Rob Tibshirani, Trevor Hastie, B. Narashiman and Gilbert Chu:
"Diagnosis of multiple cancer types by shrunken centroids of gene
expression" (PNAS website). PNAS 2002 99:65676572 (May
14). See PAM
website for software (available soon).

Bradley Efron, Trevor Hastie, Iain Johnstone and Robert Tibshirani,
Least Angle Regression
Annals of Statistics (with discussion) (2004) 32(2), 407499. A new method for variable subset selection, with the lasso and "epsilon" forward stagewise methods as special cases.
LARS Software for R and Splus.

Antoine Guisan, Thomas Edwards and Trevor Hastie
Generalized
linear and generalized additive models in studies of species
distributions: setting the scene. Ecological Modeling
(2002) 157, 89100.
2001

Trevor Hastie, Robert Tibshirani and Jerome Friedman,
"Elements
of Statistical Learning: Data Mining, Inference and
Prediction"
SpringerVerlag, New York.

Mu Zhu and Trevor Hastie,
"Feature extraction for
nonparametric discriminant analysis" JCGS (2003, 12(1), pages 101120.

Ji Zhu and Trevor Hastie,
"Kernel Logistic Regression and the Import Vector Machine", (NIPS, 2001; JCGS 2005). Copy of
slides(pdf) presented by TH in
Kyoto in December, 2001.
Ji Zhu's
R code for fitting IVM models.

Robert Tibshirani, Trevor Hastie, Balasubramanian Narasimhan,
Michael Eisen, Gavin Sherlock, Pat Brown, and David Botstein
Exploratory screening of genes and
clusters from microarray experiments (ps file) or
pdf version.

Therese Sorlie,
Perou, C., Robert Tibshirani,
Turid Aas, Stephanie Geisler,
Hilde Johnsen, Trevor Hastie,
Michael B. Eisen, Matt van de
Rijn, Stefanie S. Jeffrey,
Thor Thorsen, Hanne Quist,
John C. Matese, Patrick O.
Brown, David Botstein, Per
Eystein Lonninngg, and AnneLise BorresenDale.
Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical
implications. PNAS 98: 1086910874.
pdf version.
2000

Trevor Hastie, Robert Tibshirani, David Botstein and Pat Brown,
"Supervised Harvesting of Expression Trees" (postscript) .
Starting from a hierarchically clustered expression array, we build a
predictive model for an outcome variable using cluster nodes as inputs.
(pdf version)
Tech. report. August 2000.

Olga Troyanskaya, Michael Cantor, Gavin Sherlock,
Pat Brown, Trevor Hastie, Robert Tibshirani, David Botstein
and Russ B. Altman,
Missing value estimation methods for DNA
microarrays BIOINFORMATICS
Vol. 17 no. 6, 2001
Pages 520525

Eva Cantoni and Trevor Hastie "DegreesofFreedom Tests for Smoothing
Splines." Tech Report, May 2000.
Published in
Biometrika 2002,
89, 251263.
A mixedeffects framework for smoothing splines and additive models
allows for exact tests between nested models of different complexity.
The complexity is calibrated via the effective degrees of freedom.

Thomas Yee and Trevor Hastie.
Reduced Rank Vector Generalized Linear
Models (2003) Statistical Modeling, 3, pages 1541. Using the multinomial
as a primary example, we propose reduced rank logit models for
discrimination and classification. This is a conditional version
of the reduced rank model of linear discriminant analysis.

Robert Tibshirani, Guenther Walther and Trevor Hastie.
"Estimating the number of clusters in a dataset via the Gap statistic".
Journal of the Royal Statistical Society, B, 63:411423,2001.

Stochastic
Modeling and Tracking of Human Motion, a joint project with
Dirk Ormoneit and Michael Black's group at Xerox Parc, with
motion graphics demonstrations of learned walking
characteristics.

Page 50 of "Generalized Additive Models" by Hastie and Tibshirani,
1990, Chapman and Hall. Some copies of the 1999 printing by CRC Press
replaced page 50 with a page from a history text!
page50.ps or page50.pdf

Trevor Hastie, Laura Bachrach, Balasubramanian Narasimhan and May Choo
Wang. Flexible Statistical Models for Growth Fragments: a Study of
Bone Mineral Acquisition Compare your own measurements using our
online
growth tables

Gareth James and Trevor Hastie Functional Linear Discriminant
Analysis for Irregularly Sampled Curves (2001) Journal of the Royal
Statistical Society, Series B JRSS B 63, 533550.

Trevor Hastie, Robert Tibshirani, Michael B Eisen, Ash
Alizadeh, Ronald Levy, Louis Staudt, Wing C Chan, David Botstein,
Patrick Brown.
`Gene shaving' as a method for identifying distinct sets of genes
with similar expression patterns This is an online version
of the paper, published in the
online journal GenomeBiology.

Trevor Hastie, Robert Tibshirani, Michael Eisen, Pat Brown, Doug Ross, Uwe Scherf, John Weinstein, Ash Alizadeh, Louis Staudt, David Botstein
"Gene Shaving: a New Class of Clustering Methods for Expression
Arrays".
Postscript (2.9mb) or
Adobe pdf (5.4mb) Tech. report. Jan 2000.

James, G.,
Hastie, T., and Sugar, C. "A
Principal Component Models for Sparse Functional Data".
(2000, Biometrika, 87, 587602) (pdf). When the data are collections of sampled curves or
images, functional principal components produce the principal
modes of variation. Here we generalize these
procedures to deal with the case when each curve is sparsely and
irregularly sampled.
1999

Hastie, T., Tibshirani, R., Sherlock, G., Eisen, M., Brown, P.
and Botstein, D. "Imputing Missing
Data for Gene Expression Arrays". Technical report (1999),
Stanford Statistics Department.
pdf (145Kb) or
postscript (450Kb)

Tibshirani, R., Hastie, T. Eisen, M., Ross, D. , Botstein, D.
and Brown, P. "Clustering methods for the analysis of DNA microarray data".
Postscript (4.8mb) or
Compressed Postscript (1.8mb)
Tech. report Oct. 1999.
 D. Ormoneit and T. Hastie.
Optimal kernel shapes for local linear regression.
In S. A. Solla, T. K. Leen, and KR. Müller, editors, Advances
in Neural Information Processing Systems 12. The MIT Press, 2000.

Tibshirani, R. and Lazzeroni, L. and
Hastie, T. and Olshen, A. and Cox, D.R.
"A
Global Pairwise Approach to Radiation Hydrid Mapping".
Technical Report January 1999. Using data of cooccurrence of
hybridized markers after shattering, inference is made of the marker
sequence in the chromosome.
1998

Friedman, J., Hastie, T. and Tibshirani, R. (Published version)
Additive Logistic Regression:
a Statistical View of Boosting Annals of
Statistics 28(2), 337407. (with
discussion)
We show that boosting fits an additive logistic regression model
by stagewise optimization of a criterion very similar to the
loglikelihood, and present likelihood based alternatives. We
also propose a multilogit boosting procedure which appears to have
advantages over other methods proposed so far.
Here are the slides (2 per page) for my
boosting talk.

Crellin, N., Hastie, T. and Johnstone, I.
"Statistical Models for Image
Sequences" Technical report, submitted to "Human Brain Mapping".
We study fMRI sequences of the human brain obtained from
experiments involving repetitive neuronal activity. We investigate the
function form of
the hemodynamic response function, and provide evidence that
the commonly adopted convolution model is inadequate.

Hastie, T. and Tibshirani, R.
"Bayesian Backfitting" Stanford Technical report.
The Gibbs sampler looks and feels like the backfitting algorithm
for fitting additive models. Indeed, a simple modification to
backfitting turns it into a Gibbs sampler for spitting out
samples from the "posterior" distribution for an additive fit.
Published Statistical Science 15, no. 3 (2000), 196223

Wu, T.,Hastie, T., Schmidler, S. and Brutlag, D.
"Regression Analysis of Multiple
Protein Structures" Models for lining up and averaging
groups of protein structures.
1997

Rubinstein, D. and Hastie, T. "Discriminative vs Informative Learning" A comparison of two
frequently used but different paradigms for training classifiers.

Maes, S. and Hastie, T.
"Dynamic Mixtures of Splines: a Model for
Saliency Grouping in the Time Frequency Plane"
This is an application of mixture modeling to speech data. We
use a moving mixture of Gaussians to represent the
formantfrequencies in speech data.

Hastie, T., and Tibshirani, R. and Buja, A.
"Flexible Discriminant and Mixture Models"
In edited proceedings of "Neural Networks and
Statistics" conference, Edinburgh, 1995. J. Kay and
D. Titterington, Eds. Oxford University Press.

Wu, T., Schmidler, S., Hastie, T., and Brutlag, D.
"Modelling and superposition of
multiple protein
structures using affine transformations: analysis of the
globins"

James, G., and Hastie, T.
"Generalizations of the Bias/Variance
Decomposition for Prediction Error".
Several papers have recently appeared on this topic, and each
have a different viewpoint and decomposition. We hope ours does
not add to the confusion.

James, G., and Hastie, T.
Error Coding and PaCTs.
Gareth James' winning paper in the ASA student paper competition
for the Statistical Computing Section. This is one of four winning papers.
1996

Hastie, T.
"Neural Networks", to appear in
Encyclopaedia of Biostatistics. A brief survey with some personal
points of view.

Hastie, T., and Tibshirani, R.
"Classification by Pairwise Coupling"
We solve a multiclass classification problem by combining all the
pairwise rules. This paper builds on ideas proposed by
J. Friedman. An abbreviated
version is published in Advances in Neural Information Processing Systems 10, M. I. Jordan, M. J. Kearns, S. A. Solla, eds., MIT
Press, 1998.

Hastie, T. and Tibshirani, R.
"Generalized Additive Models" to appear in
"Encyclopaedia of Statistical Sciences". A survey paper on GAMs.
1995
 Hastie, T.
Pseudosplines.
We develop an approximation to any smoother, but typically a smoothing spline. The idea is to develop an orthonormal basis and penalty sequence, which would then be used via a generalized ridge regularization.
JRSSB(1996) 58(2), 379396.

Hastie, T. and Simard, P.
"Models and Metrics for Handwritten Character
Recognition". This paper gives a brief survey of techniques for
handwritten digit recognition, and then goes into some particular
technique based on invariant distance in some detail. Statistical Science 13(1) 1998, pp 5465.
(published version)

Hastie, T. and Tibshirani, R.
"Discriminant Adaptive Nearest Neighbor
Classification." IEEE PAMI, 18, 607616, 1996.

Hastie, T. and Tibshirani, R.
"Generalized Additive Models for Medical
Research" Encyclopaedia for Biostatistics 4 187196
1994

Hastie, T. and Tibshirani, R.
"Discriminant Analysis by Gaussian Mixtures."
JRSSB (Jan 1996). There is also a longer
technical report of Feb 1994

Hastie, T. J., Buja, A., and Tibshirani, R.
"Penalized Discriminant Analysis." (Bell Labs technical report; postscript).
Pdf of
published version, Annals of Statistics, 1995.

Hastie, T. J., Simard, P. Y., and Saeckinger, E.
"Learning Prototype Models for
Tangent Distance." NIPS proceedings, 1994.

Hastie, T.J and Tibshirani, R.
"Handwritten Digit Recognition via Deformable
Prototypes."
AT&T Bell Laboratories Technical Report, 1994.
1993

Hastie, T. J., Tibshirani, R. and Buja, A.
"Flexible Discriminant Analysis by Optimal
Scoring." (Bell Labs Technical report; postscript).
Pdf of
published version, JASA, December 1994.
 Roosen, C. B. and Hastie, T.
"Automatic Smoothing Spline Projection Pursuit."
AT&T Bell Labs Technical Report (Dec. 1993).
Journal of Computational and Graphical Statistics 3(3) Sep 1994, 235248.
Here is JSTOR
link
 Roosen, C. B. and Hastie, T.
"Logistic Response Projection Pursuit."
AT&T Bell Labs Technical Report (Aug. 1993).
1992
1990

Hastie, T. J., and Pregibon, D.
"Shrinking Trees."
AT&T Bell Laboratories Technical Report (March 1990).
Unpublished manuscript. Thanks to Mu Zhu for turning the preweb
technical memorandum into an online document.
1989

Buja, A. Hastie, T. J., Tibshirani, R.
Linear Smoothers and Additive Models
An overview of linear smoothing technology, including a proof of the
convergence of the backfitting algorithm. Annals of Statistics
(with discussion)
1989, 17(2) 453555.
 Hastie, T. and Stuetzle, W.
Principal Curves
Journal of Amarican Statistical Association 1989, 84 502516

Hastie, T., Botha, J., Schnitzler, C.
Regression with an Ordered Categorical Response
Statistics in Medicine 1989, 8 785794
1987

Hastie, T. J., and Pregibon, D.
"A
new algorithm for matched casecontrol studies with applications to
additive models."
AT&T Bell Laboratories Technical Report (November 1987).
Unpublished manuscript. A shorter version appeared in the proceedings
of Compstat 1988

Hastie, T. and Little, F.
Principal
Profiles A nonlinear version of principal components for
compositional data, that helps explain the horseshoe
effect. Published in Proceedings of the Interface Meeting (Comp.Sci
and Stat), 1987.

Greenacre, M. and Hastie, T.
A Geometric Interpretation of
Correspondence Analysis There are many ways to think of CA. This
paper presents it as a form of PCA, or subspace approximation in a
Chisquared metric. JASA (82) June 1987.
 Hastie, T.
A Closer Look at the
Deviance
The deviance for GLMs provides the analog in many ways for
sumofsquares in linear regression. This little paper surveys the
connections. The American Statistician 41(1), 1987, pp 1620
1986
1984

Principal Curves and Surfaces
SLAC (Stanford Linear Accelerator Center) has put up a pdf version of
my Ph.D thesis.
Original principal curves and surfaces movie (youtube).

Generalized Additive Models: the original technical report, written by two PhD students.
1980

