phd thesis in deep learning

Machine Learning - CMU

PhD Dissertations

[all are .pdf files].

Learning Models that Match Jacob Tyo, 2024

Improving Human Integration across the Machine Learning Pipeline Charvi Rastogi, 2024

Reliable and Practical Machine Learning for Dynamic Healthcare Settings Helen Zhou, 2023

Automatic customization of large-scale spiking network models to neuronal population activity (unavailable) Shenghao Wu, 2023

Estimation of BVk functions from scattered data (unavailable) Addison J. Hu, 2023

Rethinking object categorization in computer vision (unavailable) Jayanth Koushik, 2023

Advances in Statistical Gene Networks Jinjin Tian, 2023 Post-hoc calibration without distributional assumptions Chirag Gupta, 2023

The Role of Noise, Proxies, and Dynamics in Algorithmic Fairness Nil-Jana Akpinar, 2023

Collaborative learning by leveraging siloed data Sebastian Caldas, 2023

Modeling Epidemiological Time Series Aaron Rumack, 2023

Human-Centered Machine Learning: A Statistical and Algorithmic Perspective Leqi Liu, 2023

Uncertainty Quantification under Distribution Shifts Aleksandr Podkopaev, 2023

Probabilistic Reinforcement Learning: Using Data to Define Desired Outcomes, and Inferring How to Get There Benjamin Eysenbach, 2023

Comparing Forecasters and Abstaining Classifiers Yo Joong Choe, 2023

Using Task Driven Methods to Uncover Representations of Human Vision and Semantics Aria Yuan Wang, 2023

Data-driven Decisions - An Anomaly Detection Perspective Shubhranshu Shekhar, 2023

Applied Mathematics of the Future Kin G. Olivares, 2023

METHODS AND APPLICATIONS OF EXPLAINABLE MACHINE LEARNING Joon Sik Kim, 2023

NEURAL REASONING FOR QUESTION ANSWERING Haitian Sun, 2023

Principled Machine Learning for Societally Consequential Decision Making Amanda Coston, 2023

Long term brain dynamics extend cognitive neuroscience to timescales relevant for health and physiology Maxwell B. Wang, 2023

Long term brain dynamics extend cognitive neuroscience to timescales relevant for health and physiology Darby M. Losey, 2023

Calibrated Conditional Density Models and Predictive Inference via Local Diagnostics David Zhao, 2023

Towards an Application-based Pipeline for Explainability Gregory Plumb, 2022

Objective Criteria for Explainable Machine Learning Chih-Kuan Yeh, 2022

Making Scientific Peer Review Scientific Ivan Stelmakh, 2022

Facets of regularization in high-dimensional learning: Cross-validation, risk monotonization, and model complexity Pratik Patil, 2022

Active Robot Perception using Programmable Light Curtains Siddharth Ancha, 2022

Strategies for Black-Box and Multi-Objective Optimization Biswajit Paria, 2022

Unifying State and Policy-Level Explanations for Reinforcement Learning Nicholay Topin, 2022

Sensor Fusion Frameworks for Nowcasting Maria Jahja, 2022

Equilibrium Approaches to Modern Deep Learning Shaojie Bai, 2022

Towards General Natural Language Understanding with Probabilistic Worldbuilding Abulhair Saparov, 2022

Applications of Point Process Modeling to Spiking Neurons (Unavailable) Yu Chen, 2021

Neural variability: structure, sources, control, and data augmentation Akash Umakantha, 2021

Structure and time course of neural population activity during learning Jay Hennig, 2021

Cross-view Learning with Limited Supervision Yao-Hung Hubert Tsai, 2021

Meta Reinforcement Learning through Memory Emilio Parisotto, 2021

Learning Embodied Agents with Scalably-Supervised Reinforcement Learning Lisa Lee, 2021

Learning to Predict and Make Decisions under Distribution Shift Yifan Wu, 2021

Statistical Game Theory Arun Sai Suggala, 2021

Towards Knowledge-capable AI: Agents that See, Speak, Act and Know Kenneth Marino, 2021

Learning and Reasoning with Fast Semidefinite Programming and Mixing Methods Po-Wei Wang, 2021

Bridging Language in Machines with Language in the Brain Mariya Toneva, 2021

Curriculum Learning Otilia Stretcu, 2021

Principles of Learning in Multitask Settings: A Probabilistic Perspective Maruan Al-Shedivat, 2021

Towards Robust and Resilient Machine Learning Adarsh Prasad, 2021

Towards Training AI Agents with All Types of Experiences: A Unified ML Formalism Zhiting Hu, 2021

Building Intelligent Autonomous Navigation Agents Devendra Chaplot, 2021

Learning to See by Moving: Self-supervising 3D Scene Representations for Perception, Control, and Visual Reasoning Hsiao-Yu Fish Tung, 2021

Statistical Astrophysics: From Extrasolar Planets to the Large-scale Structure of the Universe Collin Politsch, 2020

Causal Inference with Complex Data Structures and Non-Standard Effects Kwhangho Kim, 2020

Networks, Point Processes, and Networks of Point Processes Neil Spencer, 2020

Dissecting neural variability using population recordings, network models, and neurofeedback (Unavailable) Ryan Williamson, 2020

Predicting Health and Safety: Essays in Machine Learning for Decision Support in the Public Sector Dylan Fitzpatrick, 2020

Towards a Unified Framework for Learning and Reasoning Han Zhao, 2020

Learning DAGs with Continuous Optimization Xun Zheng, 2020

Machine Learning and Multiagent Preferences Ritesh Noothigattu, 2020

Learning and Decision Making from Diverse Forms of Information Yichong Xu, 2020

Towards Data-Efficient Machine Learning Qizhe Xie, 2020

Change modeling for understanding our world and the counterfactual one(s) William Herlands, 2020

Machine Learning in High-Stakes Settings: Risks and Opportunities Maria De-Arteaga, 2020

Data Decomposition for Constrained Visual Learning Calvin Murdock, 2020

Structured Sparse Regression Methods for Learning from High-Dimensional Genomic Data Micol Marchetti-Bowick, 2020

Towards Efficient Automated Machine Learning Liam Li, 2020

LEARNING COLLECTIONS OF FUNCTIONS Emmanouil Antonios Platanios, 2020

Provable, structured, and efficient methods for robustness of deep networks to adversarial examples Eric Wong , 2020

Reconstructing and Mining Signals: Algorithms and Applications Hyun Ah Song, 2020

Probabilistic Single Cell Lineage Tracing Chieh Lin, 2020

Graphical network modeling of phase coupling in brain activity (unavailable) Josue Orellana, 2019

Strategic Exploration in Reinforcement Learning - New Algorithms and Learning Guarantees Christoph Dann, 2019 Learning Generative Models using Transformations Chun-Liang Li, 2019

Estimating Probability Distributions and their Properties Shashank Singh, 2019

Post-Inference Methods for Scalable Probabilistic Modeling and Sequential Decision Making Willie Neiswanger, 2019

Accelerating Text-as-Data Research in Computational Social Science Dallas Card, 2019

Multi-view Relationships for Analytics and Inference Eric Lei, 2019

Information flow in networks based on nonstationary multivariate neural recordings Natalie Klein, 2019

Competitive Analysis for Machine Learning & Data Science Michael Spece, 2019

The When, Where and Why of Human Memory Retrieval Qiong Zhang, 2019

Towards Effective and Efficient Learning at Scale Adams Wei Yu, 2019

Towards Literate Artificial Intelligence Mrinmaya Sachan, 2019

Learning Gene Networks Underlying Clinical Phenotypes Under SNP Perturbations From Genome-Wide Data Calvin McCarter, 2019

Unified Models for Dynamical Systems Carlton Downey, 2019

Anytime Prediction and Learning for the Balance between Computation and Accuracy Hanzhang Hu, 2019

Statistical and Computational Properties of Some "User-Friendly" Methods for High-Dimensional Estimation Alnur Ali, 2019

Nonparametric Methods with Total Variation Type Regularization Veeranjaneyulu Sadhanala, 2019

New Advances in Sparse Learning, Deep Networks, and Adversarial Learning: Theory and Applications Hongyang Zhang, 2019

Gradient Descent for Non-convex Problems in Modern Machine Learning Simon Shaolei Du, 2019

Selective Data Acquisition in Learning and Decision Making Problems Yining Wang, 2019

Anomaly Detection in Graphs and Time Series: Algorithms and Applications Bryan Hooi, 2019

Neural dynamics and interactions in the human ventral visual pathway Yuanning Li, 2018

Tuning Hyperparameters without Grad Students: Scaling up Bandit Optimisation Kirthevasan Kandasamy, 2018

Teaching Machines to Classify from Natural Language Interactions Shashank Srivastava, 2018

Statistical Inference for Geometric Data Jisu Kim, 2018

Representation Learning @ Scale Manzil Zaheer, 2018

Diversity-promoting and Large-scale Machine Learning for Healthcare Pengtao Xie, 2018

Distribution and Histogram (DIsH) Learning Junier Oliva, 2018

Stress Detection for Keystroke Dynamics Shing-Hon Lau, 2018

Sublinear-Time Learning and Inference for High-Dimensional Models Enxu Yan, 2018

Neural population activity in the visual cortex: Statistical methods and application Benjamin Cowley, 2018

Efficient Methods for Prediction and Control in Partially Observable Environments Ahmed Hefny, 2018

Learning with Staleness Wei Dai, 2018

Statistical Approach for Functionally Validating Transcription Factor Bindings Using Population SNP and Gene Expression Data Jing Xiang, 2017

New Paradigms and Optimality Guarantees in Statistical Learning and Estimation Yu-Xiang Wang, 2017

Dynamic Question Ordering: Obtaining Useful Information While Reducing User Burden Kirstin Early, 2017

New Optimization Methods for Modern Machine Learning Sashank J. Reddi, 2017

Active Search with Complex Actions and Rewards Yifei Ma, 2017

Why Machine Learning Works George D. Montañez , 2017

Source-Space Analyses in MEG/EEG and Applications to Explore Spatio-temporal Neural Dynamics in Human Vision Ying Yang , 2017

Computational Tools for Identification and Analysis of Neuronal Population Activity Pengcheng Zhou, 2016

Expressive Collaborative Music Performance via Machine Learning Gus (Guangyu) Xia, 2016

Supervision Beyond Manual Annotations for Learning Visual Representations Carl Doersch, 2016

Exploring Weakly Labeled Data Across the Noise-Bias Spectrum Robert W. H. Fisher, 2016

Optimizing Optimization: Scalable Convex Programming with Proximal Operators Matt Wytock, 2016

Combining Neural Population Recordings: Theory and Application William Bishop, 2015

Discovering Compact and Informative Structures through Data Partitioning Madalina Fiterau-Brostean, 2015

Machine Learning in Space and Time Seth R. Flaxman, 2015

The Time and Location of Natural Reading Processes in the Brain Leila Wehbe, 2015

Shape-Constrained Estimation in High Dimensions Min Xu, 2015

Spectral Probabilistic Modeling and Applications to Natural Language Processing Ankur Parikh, 2015 Computational and Statistical Advances in Testing and Learning Aaditya Kumar Ramdas, 2015

Corpora and Cognition: The Semantic Composition of Adjectives and Nouns in the Human Brain Alona Fyshe, 2015

Learning Statistical Features of Scene Images Wooyoung Lee, 2014

Towards Scalable Analysis of Images and Videos Bin Zhao, 2014

Statistical Text Analysis for Social Science Brendan T. O'Connor, 2014

Modeling Large Social Networks in Context Qirong Ho, 2014

Semi-Cooperative Learning in Smart Grid Agents Prashant P. Reddy, 2013

On Learning from Collective Data Liang Xiong, 2013

Exploiting Non-sequence Data in Dynamic Model Learning Tzu-Kuo Huang, 2013

Mathematical Theories of Interaction with Oracles Liu Yang, 2013

Short-Sighted Probabilistic Planning Felipe W. Trevizan, 2013

Statistical Models and Algorithms for Studying Hand and Finger Kinematics and their Neural Mechanisms Lucia Castellanos, 2013

Approximation Algorithms and New Models for Clustering and Learning Pranjal Awasthi, 2013

Uncovering Structure in High-Dimensions: Networks and Multi-task Learning Problems Mladen Kolar, 2013

Learning with Sparsity: Structures, Optimization and Applications Xi Chen, 2013

GraphLab: A Distributed Abstraction for Large Scale Machine Learning Yucheng Low, 2013

Graph Structured Normal Means Inference James Sharpnack, 2013 (Joint Statistics & ML PhD)

Probabilistic Models for Collecting, Analyzing, and Modeling Expression Data Hai-Son Phuoc Le, 2013

Learning Large-Scale Conditional Random Fields Joseph K. Bradley, 2013

New Statistical Applications for Differential Privacy Rob Hall, 2013 (Joint Statistics & ML PhD)

Parallel and Distributed Systems for Probabilistic Reasoning Joseph Gonzalez, 2012

Spectral Approaches to Learning Predictive Representations Byron Boots, 2012

Attribute Learning using Joint Human and Machine Computation Edith L. M. Law, 2012

Statistical Methods for Studying Genetic Variation in Populations Suyash Shringarpure, 2012

Data Mining Meets HCI: Making Sense of Large Graphs Duen Horng (Polo) Chau, 2012

Learning with Limited Supervision by Input and Output Coding Yi Zhang, 2012

Target Sequence Clustering Benjamin Shih, 2011

Nonparametric Learning in High Dimensions Han Liu, 2010 (Joint Statistics & ML PhD)

Structural Analysis of Large Networks: Observations and Applications Mary McGlohon, 2010

Modeling Purposeful Adaptive Behavior with the Principle of Maximum Causal Entropy Brian D. Ziebart, 2010

Tractable Algorithms for Proximity Search on Large Graphs Purnamrita Sarkar, 2010

Rare Category Analysis Jingrui He, 2010

Coupled Semi-Supervised Learning Andrew Carlson, 2010

Fast Algorithms for Querying and Mining Large Graphs Hanghang Tong, 2009

Efficient Matrix Models for Relational Learning Ajit Paul Singh, 2009

Exploiting Domain and Task Regularities for Robust Named Entity Recognition Andrew O. Arnold, 2009

Theoretical Foundations of Active Learning Steve Hanneke, 2009

Generalized Learning Factors Analysis: Improving Cognitive Models with Machine Learning Hao Cen, 2009

Detecting Patterns of Anomalies Kaustav Das, 2009

Dynamics of Large Networks Jurij Leskovec, 2008

Computational Methods for Analyzing and Modeling Gene Regulation Dynamics Jason Ernst, 2008

Stacked Graphical Learning Zhenzhen Kou, 2007

Actively Learning Specific Function Properties with Applications to Statistical Inference Brent Bryan, 2007

Approximate Inference, Structure Learning and Feature Estimation in Markov Random Fields Pradeep Ravikumar, 2007

Scalable Graphical Models for Social Networks Anna Goldenberg, 2007

Measure Concentration of Strongly Mixing Processes with Applications Leonid Kontorovich, 2007

Tools for Graph Mining Deepayan Chakrabarti, 2005

Automatic Discovery of Latent Variable Models Ricardo Silva, 2005

• Warning : Invalid argument supplied for foreach() in /home/customer/www/opendatascience.com/public_html/wp-includes/nav-menu.php on line 95 Warning : array_merge(): Expected parameter 2 to be an array, null given in /home/customer/www/opendatascience.com/public_html/wp-includes/nav-menu.php on line 102
• ODSC EUROPE
• AI+ Training
• Speak at ODSC

• Data Analytics
• Data Engineering
• Data Visualization
• Deep Learning
• Generative AI
• Machine Learning
• NLP and LLMs
• Business & Use Cases
• Career Advice
• Write for us
• ODSC Community Slack Channel
• Upcoming Webinars

10 Compelling Machine Learning Ph.D. Dissertations for 2020

Machine Learning Modeling Research posted by Daniel Gutierrez, ODSC August 19, 2020 Daniel Gutierrez, ODSC

As a data scientist, an integral part of my work in the field revolves around keeping current with research coming out of academia. I frequently scour arXiv.org for late-breaking papers that show trends and reveal fertile areas of research. Other sources of valuable research developments are in the form of Ph.D. dissertations, the culmination of a doctoral candidate’s work to confer his/her degree. Ph.D. candidates are highly motivated to choose research topics that establish new and creative paths toward discovery in their field of study. Their dissertations are highly focused on a specific problem. If you can find a dissertation that aligns with your areas of interest, consuming the research is an excellent way to do a deep dive into the technology. After reviewing hundreds of recent theses from universities all over the country, I present 10 machine learning dissertations that I found compelling in terms of my own areas of interest.

[Related article: Introduction to Bayesian Deep Learning ]

I hope you’ll find several that match your own fields of inquiry. Each thesis may take a while to consume but will result in hours of satisfying summer reading. Enjoy!

1. Bayesian Modeling and Variable Selection for Complex Data

As we routinely encounter high-throughput data sets in complex biological and environmental research, developing novel models and methods for variable selection has received widespread attention. This dissertation addresses a few key challenges in Bayesian modeling and variable selection for high-dimensional data with complex spatial structures. 

2. Topics in Statistical Learning with a Focus on Large Scale Data

Big data vary in shape and call for different approaches. One type of big data is the tall data, i.e., a very large number of samples but not too many features. This dissertation describes a general communication-efficient algorithm for distributed statistical learning on this type of big data. The algorithm distributes the samples uniformly to multiple machines, and uses a common reference data to improve the performance of local estimates. The algorithm enables potentially much faster analysis, at a small cost to statistical performance.

Another type of big data is the wide data, i.e., too many features but a limited number of samples. It is also called high-dimensional data, to which many classical statistical methods are not applicable. 

This dissertation discusses a method of dimensionality reduction for high-dimensional classification. The method partitions features into independent communities and splits the original classification problem into separate smaller ones. It enables parallel computing and produces more interpretable results.

3. Sets as Measures: Optimization and Machine Learning

The purpose of this machine learning dissertation is to address the following simple question:

How do we design efficient algorithms to solve optimization or machine learning problems where the decision variable (or target label) is a set of unknown cardinality?

Optimization and machine learning have proved remarkably successful in applications requiring the choice of single vectors. Some tasks, in particular many inverse problems, call for the design, or estimation, of sets of objects. When the size of these sets is a priori unknown, directly applying optimization or machine learning techniques designed for single vectors appears difficult. The work in this dissertation shows that a very old idea for transforming sets into elements of a vector space (namely, a space of measures), a common trick in theoretical analysis, generates effective practical algorithms.

4. A Geometric Perspective on Some Topics in Statistical Learning

Modern science and engineering often generate data sets with a large sample size and a comparably large dimension which puts classic asymptotic theory into question in many ways. Therefore, the main focus of this dissertation is to develop a fundamental understanding of statistical procedures for estimation and hypothesis testing from a non-asymptotic point of view, where both the sample size and problem dimension grow hand in hand. A range of different problems are explored in this thesis, including work on the geometry of hypothesis testing, adaptivity to local structure in estimation, effective methods for shape-constrained problems, and early stopping with boosting algorithms. The treatment of these different problems shares the common theme of emphasizing the underlying geometric structure.

5. Essays on Random Forest Ensembles

A random forest is a popular machine learning ensemble method that has proven successful in solving a wide range of classification problems. While other successful classifiers, such as boosting algorithms or neural networks, admit natural interpretations as maximum likelihood, a suitable statistical interpretation is much more elusive for a random forest. The first part of this dissertation demonstrates that a random forest is a fruitful framework in which to study AdaBoost and deep neural networks. The work explores the concept and utility of interpolation, the ability of a classifier to perfectly fit its training data. The second part of this dissertation places a random forest on more sound statistical footing by framing it as kernel regression with the proximity kernel. The work then analyzes the parameters that control the bandwidth of this kernel and discuss useful generalizations.

6. Marginally Interpretable Generalized Linear Mixed Models

A popular approach for relating correlated measurements of a non-Gaussian response variable to a set of predictors is to introduce latent random variables and fit a generalized linear mixed model. The conventional strategy for specifying such a model leads to parameter estimates that must be interpreted conditional on the latent variables. In many cases, interest lies not in these conditional parameters, but rather in marginal parameters that summarize the average effect of the predictors across the entire population. Due to the structure of the generalized linear mixed model, the average effect across all individuals in a population is generally not the same as the effect for an average individual. Further complicating matters, obtaining marginal summaries from a generalized linear mixed model often requires evaluation of an analytically intractable integral or use of an approximation. Another popular approach in this setting is to fit a marginal model using generalized estimating equations. This strategy is effective for estimating marginal parameters, but leaves one without a formal model for the data with which to assess quality of fit or make predictions for future observations. Thus, there exists a need for a better approach.

This dissertation defines a class of marginally interpretable generalized linear mixed models that leads to parameter estimates with a marginal interpretation while maintaining the desirable statistical properties of a conditionally specified model. The distinguishing feature of these models is an additive adjustment that accounts for the curvature of the link function and thereby preserves a specific form for the marginal mean after integrating out the latent random variables. 

7. On the Detection of Hate Speech, Hate Speakers and Polarized Groups in Online Social Media

The objective of this dissertation is to explore the use of machine learning algorithms in understanding and detecting hate speech, hate speakers and polarized groups in online social media. Beginning with a unique typology for detecting abusive language, the work outlines the distinctions and similarities of different abusive language subtasks (offensive language, hate speech, cyberbullying and trolling) and how we might benefit from the progress made in each area. Specifically, the work suggests that each subtask can be categorized based on whether or not the abusive language being studied 1) is directed at a specific individual, or targets a generalized “Other” and 2) the extent to which the language is explicit versus implicit. The work then uses knowledge gained from this typology to tackle the “problem of offensive language” in hate speech detection. 

8. Lasso Guarantees for Dependent Data

Serially correlated high dimensional data are prevalent in the big data era. In order to predict and learn the complex relationship among the multiple time series, high dimensional modeling has gained importance in various fields such as control theory, statistics, economics, finance, genetics and neuroscience. This dissertation studies a number of high dimensional statistical problems involving different classes of mixing processes. 

9. Random forest robustness, variable importance, and tree aggregation

Random forest methodology is a nonparametric, machine learning approach capable of strong performance in regression and classification problems involving complex data sets. In addition to making predictions, random forests can be used to assess the relative importance of feature variables. This dissertation explores three topics related to random forests: tree aggregation, variable importance, and robustness. 

10. Climate Data Computing: Optimal Interpolation, Averaging, Visualization and Delivery

This dissertation solves two important problems in the modern analysis of big climate data. The first is the efficient visualization and fast delivery of big climate data, and the second is a computationally extensive principal component analysis (PCA) using spherical harmonics on the Earth’s surface. The second problem creates a way to supply the data for the technology developed in the first. These two problems are computationally difficult, such as the representation of higher order spherical harmonics Y400, which is critical for upscaling weather data to almost infinitely fine spatial resolution.

I hope you enjoyed learning about these compelling machine learning dissertations.

Editor’s note: Interested in more data science research? Check out the Research Frontiers track at ODSC Europe this September 17-19 or the ODSC West Research Frontiers track this October 27-30.

Daniel Gutierrez, ODSC

Daniel D. Gutierrez is a practicing data scientist who’s been working with data long before the field came in vogue. As a technology journalist, he enjoys keeping a pulse on this fast-paced industry. Daniel is also an educator having taught data science, machine learning and R classes at the university level. He has authored four computer industry books on database and data science technology, including his most recent title, “Machine Learning and Data Science: An Introduction to Statistical Learning Methods with R.” Daniel holds a BS in Mathematics and Computer Science from UCLA.

Meta Sees Free Models as its Future

AI and Data Science News posted by ODSC Team Apr 27, 2024 A few days ago, Meta introduced Llama 3, its latest advanced AI model, to the public...

ODSC’s AI Weekly Recap: Week of April 26th

AI and Data Science News posted by Jorge Arenas Apr 26, 2024 Every week, the ODSC team researches the latest advancements in AI. We review a selection of...

New AI Models From Apple May Find Home in Future iPhones

AI and Data Science News posted by ODSC Team Apr 25, 2024 In a report from the Independent AI, Apple researchers have introduced a series of new AI...

MIT Libraries home DSpace@MIT

• DSpace@MIT Home
• MIT Libraries
• Doctoral Theses

On optimization and scalability in deep learning

Other Contributors

Terms of use, description, date issued, collections.

• Faculty of Arts and Sciences
• FAS Theses and Dissertations
• Communities & Collections
• By Issue Date
• FAS Department
• Quick submit
• Waiver Generator
• DASH Stories
• Accessibility
• COVID-related Research

Terms of Use

• Privacy Policy
• By Collections
• By Departments

Enabling High Performance, Efficient, and Sustainable Deep Learning Systems At Scale

Citable link to this page

Collections.

• FAS Theses and Dissertations [6136]

Contact administrator regarding this item (to report mistakes or request changes)

Uncertainty in Deep Learning (PhD Thesis)

October 13th, 2016 (Updated: June 4th, 2017)

Tweet Share

Function draws from a dropout neural network. This new visualisation technique depicts the distribution over functions rather than the predictive distribution (see demo below ).

Thesis: uncertainty in deep learning.

• some discussions : a discussion of AI safety and model uncertainty ( §1.3 ), a historical survey of Bayesian neural networks ( §2.2 ),
• some theoretical analysis : a theoretical analysis of the variance of the re-parametrisation trick and other Monte Carlo estimators used in variational inference (the re-parametrisation trick is not a universal variance reduction technique! §3.1.1 – §3.1.2 ), a survey of measures of uncertainty in classification tasks ( §3.3.1 ),
• some empirical results : an empirical analysis of different Bayesian neural network priors ( §4.1 ) and posteriors with various approximating distributions ( §4.2 ), new quantitative results comparing dropout to existing techniques ( §4.3 ), tools for heteroscedastic model uncertainty in Bayesian neural networks ( §4.6 ),
• some applications : a survey of recent applications in language, biology, medicine, and computer vision making use of the tools presented in this thesis ( §5.1 ), new applications in active learning with image data ( §5.2 ),
• and more theoretical results : a discussion of what determines what our model uncertainty looks like ( §6.1 – §6.2 ), an analytical analysis of the dropout approximating distribution in Bayesian linear regression ( §6.3 ), an analysis of ELBO-test log likelihood correlation ( §6.4 ), discrete prior models ( §6.5 ), an interpretation of dropout as a proxy posterior in spike and slab prior models ( §6.6 , relating dropout to works by MacKay, Nowlan, and Hinton from 1992), as well as a procedure to optimise the dropout probabilities based on the variational interpretation to separate the different sources of uncertainty ( §6.7 ).
• Contents ( PDF , 36K)
• Chapter 1: The Importance of Knowing What We Don't Know ( PDF , 393K)
• Chapter 2: The Language of Uncertainty ( PDF , 136K)
• Chapter 3: Bayesian Deep Learning ( PDF , 302K)
• Chapter 4: Uncertainty Quality ( PDF , 2.9M)
• Chapter 5: Applications ( PDF , 648K)
• Chapter 6: Deep Insights ( PDF , 939K)
• Chapter 7: Future Research ( PDF , 28K)
• Bibliography ( PDF , 72K)
• Appendix A: KL condition ( PDF , 71K)
• Appendix B: Figures ( PDF , 2M)
• Appendix C: Spike and slab prior KL ( PDF , 28K)

Function visualisation

$ \newcommand{\N}{\mathcal{N}} \newcommand{\x}{\mathbf{x}} \newcommand{\y}{\mathbf{y}} \newcommand{\X}{\mathbf{X}} \newcommand{\Y}{\mathbf{Y}} \newcommand{\train}{{\text{train}}} \newcommand{\bo}{\text{$\omega$}} \newcommand{\yh}{\hat{\y}} \newcommand{\boh}{\text{$\hat{\omega}$}} \newcommand{\liminfT}{\xrightarrow[T \to \infty]{}} $

There are two factors at play when visualising uncertainty in dropout Bayesian neural networks: the dropout masks and the dropout probability of the first layer. Uncertainty depictions in my previous blog posts drew new dropout masks for each test point—which is equivalent to drawing a new prediction from the predictive distribution for each test point $-2 \leq \x \leq 2$. More specifically, for each test point $\x_i$ we drew a set of network parameters from the dropout approximate posterior $\boh_{i} \sim q_\theta(\bo)$, and conditioned on these parameters we drew a prediction from the likelihood $\y_i \sim p(\y | \x_i, \boh_{i})$. Since the predictive distribution has \begin{align*} p&(\y_i | \x_i, \X_\train, \Y_\train) \\ &= \int p(\y_i | \x_i, \bo) p(\bo | \X_\train, \Y_\train) \text{d} \bo \\ &\approx \int p(\y_i | \x_i, \bo) q_\theta(\bo) \text{d} \bo \\ &=: q_\theta(\y_i | \x_i) \end{align*} we have that $\y_i$ is a draw from an approximation to the predictive distribution.

Figure A: In black is a draw from the predictive distribution of a dropout neural network $\yh \sim q_\theta(\y | \x)$ for each test point $-2 \leq \x \leq 2$, compared to the function draws in figure B.

Figure B: Each solid black line is a function drawn from a dropout neural network posterior over functions, induced by a draw from the approximate posterior over the weights $\boh \sim q_\theta(\bo)$.

Another important factor affecting visualisation is the dropout probability of the first layer. In the previous posts we depicted scalar functions and set all dropout probabilities to $0.1$. As a result, with probability $0.1$, the sampled functions from the posterior would be identically zero. This is because a zero draw from the Bernoulli distribution in the first layer together with a scalar input leads the model to completely drop its input (explaining the points where the function touches the $x$-axis in figure A). This is a behaviour we might not believe the posterior should exhibit (especially when a single set of masks is drawn for the entire test set), and could change this by setting a different probability for the first layer. Setting $p_1 = 0$ for example is identical to placing a delta approximating distribution over the first weight layer.

In the demo below we use $p_1 = 0$, and depict draws from the approximate predictive distribution evaluated on the entire test set $q_{\theta_i}(\Y | \X, \boh_i)$ ($\boh_i \sim q_{\theta_i}(\bo)$, function draws ), as the variational parameters $\theta_i$ change and adapt to minimise the divergence to the true posterior (with old samples disappearing after 20 optimisation steps). You can change the function the data is drawn from (with two functions, one from the last blog post and one from the appendix in this paper ), and the model used (a homoscedastic model or a heteroscedastic model, see section §4.6 in the thesis for example or this blog post ).

• Show Uncertainty
• Heteroscedastic
• $y = x \sin x$
• $y = x + \sin 4 (x + \epsilon) + \sin 13 (x + \epsilon) + \epsilon$, $\epsilon \sim \N(0, 0.03^2)$

Last 20 function draws (together with predictive mean and predictive variance) for a dropout neural network , as the approximate posterior is transformed to fit the true posterior. This is done for two functions, both for a homoscedastic model as well as for a heteroscedastic one. You can add new points by left clicking.

Acknowledgements.

To finish this blog post I would like to thank the people that helped through comments and discussions during the writing of the various papers composing the thesis above. I would like to thank (in alphabetical order) Christof Angermueller, Yoshua Bengio, Phil Blunsom, Yutian Chen, Roger Frigola, Shane Gu, Alex Kendall, Yingzhen Li, Rowan McAllister, Carl Rasmussen, Ilya Sutskever, Gabriel Synnaeve, Nilesh Tripuraneni, Richard Turner, Oriol Vinyals, Adrian Weller, Mark van der Wilk, Yan Wu, and many other reviewers for their helpful comments and discussions. I would further like to thank my collaborators Rowan McAllister, Carl Rasmussen, Richard Turner, Mark van der Wilk, and my supervisor Zoubin Ghahramani.

Lastly, I would like to thank Google for supporting three years of my PhD with the Google European Doctoral Fellowship in Machine Learning, and Qualcomm for supporting my fourth year with the Qualcomm Innovation Fellowship.

PS. there might be some easter eggs hidden in the introduction :)

PhD Thesis: Geometry and Uncertainty in Deep Learning for Computer Vision

Chaoqiang Zhao

Monocular Depth Estimation, VO

• University of Bologna
• Google Scholar

Today I can share my final PhD thesis, which I submitted in November 2017. It was examined by Dr. Joan Lasenby and Prof. Andrew Zisserman in February 2018 and has just been approved for publication. This thesis presents the main narrative of my research at the University of Cambridge, under the supervision of Prof Roberto Cipolla. It contains 206 pages, 62 figures, 24 tables and 318 citations. You can download the complete .pdf here .

My thesis presents contributions to the field of computer vision, the science which enables machines to see. This blog post introduces the work and tells the story behind this research.

This thesis presents deep learning models for an array of computer vision problems: semantic segmentation , instance segmentation , depth prediction , localisation , stereo vision and video scene understanding .

The abstract

Deep learning and convolutional neural networks have become the dominant tool for computer vision. These techniques excel at learning complicated representations from data using supervised learning. In particular, image recognition models now out-perform human baselines under constrained settings. However, the science of computer vision aims to build machines which can see. This requires models which can extract richer information than recognition, from images and video. In general, applying these deep learning models from recognition to other problems in computer vision is significantly more challenging.

This thesis presents end-to-end deep learning architectures for a number of core computer vision problems; scene understanding, camera pose estimation, stereo vision and video semantic segmentation. Our models outperform traditional approaches and advance state-of-the-art on a number of challenging computer vision benchmarks. However, these end-to-end models are often not interpretable and require enormous quantities of training data.

To address this, we make two observations: (i) we do not need to learn everything from scratch, we know a lot about the physical world, and (ii) we cannot know everything from data, our models should be aware of what they do not know. This thesis explores these ideas using concepts from geometry and uncertainty. Specifically, we show how to improve end-to-end deep learning models by leveraging the underlying geometry of the problem. We explicitly model concepts such as epipolar geometry to learn with unsupervised learning, which improves performance. Secondly, we introduce ideas from probabilistic modelling and Bayesian deep learning to understand uncertainty in computer vision models. We show how to quantify different types of uncertainty, improving safety for real world applications.

I began my PhD in October 2014, joining the controls research group at Cambridge University Engineering Department. Looking back at my original research proposal, I said that I wanted to work on the ‘engineering questions to control autonomous vehicles… in uncertain and challenging environments.’ I spent three months or so reading literature, and quickly developed the opinion that the field of robotics was most limited by perception. If you could obtain a reliable state of the world, control was often simple. However, at this time, computer vision was very fragile in the wild. After many weeks of lobbying Prof. Roberto Cipolla (thanks!), I was able to join his research group in January 2015 and begin a PhD in computer vision.

When I began reading computer vision literature, deep learning had just become popular in image classification, following inspiring breakthroughs on the ImageNet dataset. But it was yet to become ubiquitous in the field and be used in richer computer vision tasks such as scene understanding. What excited me about deep learning was that it could learn representations from data that are too complicated to hand-design.

I initially focused on building end-to-end deep learning models for computer vision tasks which I thought were most interesting for robotics, such as scene understanding (SegNet) and localisation (PoseNet) . However, I quickly realised that, while it was a start, applying end-to-end deep learning wasn’t enough. In my thesis, I argue that we can do better than naive end-to-end convolutional networks. Especially with limited data and compute, we can form more powerful computer vision models by leveraging our knowledge of the world. Specifically, I focus on two ideas around geometry and uncertainty.

• Geometry is all about leveraging structure of the world. This is useful for improving architectures and learning with self-supervision.
• Uncertainty understands what our model doesn’t know. This is useful for robust learning, safety-critical systems and active learning.

Over the last three years, I have had the pleasure of working with some incredibly talented researchers, studying a number of core computer vision problems from localisation to segmentation to stereo vision.

The science

This thesis consists of six chapters. Each of the main chapters introduces an end-to-end deep learning model and discusses how to apply the ideas of geometry and uncertainty.

Chatper 1 - Introduction. Motivates this work within the wider field of computer vision.

Chapter 2 - Scene Understanding. Introduces SegNet, modelling aleatoric and epistemic uncertainty and a method for learning multi-task scene understanding models for geometry and semantics.

Chapter 3 - Localisation. Describes PoseNet for efficient localisation, with improvements using geometric reprojection error and estimating relocalisation uncertainty.

Chapter 4 - Stereo Vision. Designs an end-to-end model for stereo vision, using geometry and shows how to leverage uncertainty and self-supervised learning to improve performance.

Chapter 5 - Video Scene Understanding. Illustrates a video scene understanding model for learning semantics, motion and geometry.

Chapter 6 - Conclusions. Describes limitations of this research and future challenges.

As for what’s next?

This thesis explains how to extract a robust state of the world – semantics, motion and geometry – from video. I’m now excited about applying these ideas to robotics and learning to reason from perception to action. I’m working with an amazing team on autonomous driving, bringing together the worlds of robotics and machine learning. We’re using ideas from computer vision and reinforcement learning to build the most data-efficient self-driving car. And, we’re hiring, come work with me! wayve.ai/careers

I’d like to give a huge thank you to everyone who motivated, distracted and inspired me while writing this thesis.

Here’s the bibtex if you’d like to cite this work.

And the source code for the latex document is here .

• Help & FAQ

PhD thesis: Foundations and advances in deep learning

• Computer Science

Research output : Book/Report › Other report

T1 - PhD thesis

T2 - Foundations and advances in deep learning

AU - Cho, Kyunghyun

M3 - Other report

BT - PhD thesis

PB - Aalto University

Probabilistic Machine Learning in the Age of Deep Learning: New Perspectives for Gaussian Processes, Bayesian Optimization and Beyond (PhD Thesis)

Advances in artificial intelligence (AI) are rapidly transforming our world, with systems now matching or surpassing human capabilities in areas ranging from game-playing to scientific discovery. Much of this progress traces back to machine learning (ML), particularly deep learning and its ability to uncover meaningful patterns and representations in data. However, true intelligence in AI demands more than raw predictive power; it requires a principled approach to making decisions under uncertainty. This highlights the necessity of probabilistic ML, which offers a systematic framework for reasoning about the unknown through probability theory and Bayesian inference. Gaussian processes (GPs) stand out as a quintessential probabilistic model, offering flexibility, data efficiency, and well-calibrated uncertainty estimates. They are integral to many sequential decision-making algorithms, notably Bayesian optimisation (BO), which has emerged as an indispensable tool for optimising expensive and complex black-box objective functions. While considerable efforts have focused on improving gp scalability, performance gaps persist in practice when compared against neural networks (NNs) due in large to its lack of representation learning capabilities. This, among other natural deficiencies of GPs, has hampered the capacity of BO to address critical real-world optimisation challenges. This thesis aims to unlock the potential of deep learning within probabilistic methods and reciprocally lend probabilistic perspectives to deep learning. The contributions include improving approximations to bridge the gap between GPs and NNs, providing a new formulation of BO that seamlessly accommodates deep learning methods to tackle complex optimisation problems, as well as a probabilistic interpretation of a powerful class of deep generative models for image style transfer. By enriching the interplay between deep learning and probabilistic ML, this thesis advances the foundations of AI and facilitates the development of more capable and dependable automated decision-making systems.

The full text is available as a single PDF file

You can also find a list of contents and PDFs corresponding to each individual chapter below:

Table of Contents

• Chapter 1: Introduction
• Chapter 2: Background
• Chapter 3: Orthogonally-Decoupled Sparse Gaussian Processes with Spherical Neural Network Activation Features
• Chapter 4: Cycle-Consistent Generative Adversarial Networks as a Bayesian Approximation
• Chapter 5: Bayesian Optimisation by Classification with Deep Learning and Beyond
• Chapter 6: Conclusion
• Appendix A: Numerical Methods for Improved Decoupled Sampling of Gaussian Processes
• Bibliography

Please find Chapter 1: Introduction reproduced in full below:

Introduction

PhD Candidate (AI & Machine Learning)

Thanks for stopping by! Let’s connect – drop me a message or follow me

DAVID STUTZ

Phd thesis: uncertainty and robustness.

• Download and Citing

Defense Talk

Quick links: Thesis | Defense | Thesis Template | Defense Template

Figure 1: Problems tackled in my PhD thesis, ranging from 3D reconstruction in the context of autonomous driving to adversarial robustness, bit error robustness and uncertainty estimation.

Deep learning is becoming increasingly relevant for many high-stakes applications such as autonomous driving or medical diagnosis where wrong decisions can have massive impact on human lives. Unfortunately, deep neural networks are typically assessed solely based on generalization, e.g., accuracy on a fixed test set. However, this is clearly insufficient for safe deployment as potential malicious actors and distribution shifts or the effects of quantization and unreliable hardware are disregarded. Thus, recent work additionally evaluates performance on potentially manipulated or corrupted inputs as well as after quantization and deployment on specialized hardware. In such settings, it is also important to obtain reasonable estimates of the model's confidence alongside its predictions. This thesis studies robustness and uncertainty estimation in deep learning along three main directions: First, we consider so-called adversarial examples, slightly perturbed inputs causing severe drops in accuracy. Second, we study weight perturbations, focusing particularly on bit errors in quantized weights. This is relevant for deploying models on special-purpose hardware for efficient inference, so-called accelerators. Finally, we address uncertainty estimation to improve robustness and provide meaningful statistical performance guarantees for safe deployment.

In detail, we study the existence of adversarial examples with respect to the underlying data manifold. In this context, we also investigate adversarial training which improves robustness by augmenting training with adversarial examples at the cost of reduced accuracy. We show that regular adversarial examples leave the data manifold in an almost orthogonal direction. While we find no inherent trade-off between robustness and accuracy, this contributes to a higher sample complexity as well as severe overfitting of adversarial training. Using a novel measure of flatness in the robust loss landscape with respect to weight changes, we also show that robust overfitting is caused by converging to particularly sharp minima. In fact, we find a clear correlation between flatness and good robust generalization.

Further, we study random and adversarial bit errors in quantized weights. In accelerators, random bit errors occur in the memory when reducing voltage with the goal of improving energy-efficiency. Here, we consider a robust quantization scheme, use weight clipping as regularization and perform random bit error training to improve bit error robustness, allowing considerable energy savings without requiring hardware changes. In contrast, adversarial bit errors are maliciously introduced through hardware- or software-based attacks on the memory, with severe consequences on performance. We propose a novel adversarial bit error attack to study this threat and use adversarial bit error training to improve robustness and thereby also the accelerator's security.

Finally, we view robustness in the context of uncertainty estimation. By encouraging low-confidence predictions on adversarial examples, our confidence-calibrated adversarial training successfully rejects adversarial, corrupted as well as out-of-distribution examples at test time. Thereby, we are also able to improve the robustness-accuracy trade-off compared to regular adversarial training. However, even robust models do not provide any guarantee for safe deployment. To address this problem, conformal prediction allows the model to predict confidence sets with user-specified guarantee of including the true label. Unfortunately, as conformal prediction is usually applied after training, the model is trained without taking this calibration step into account. To address this limitation, we propose conformal training which allows training conformal predictors end-to-end with the underlying model. This not only improves the obtained uncertainty estimates but also enables optimizing application-specific objectives without losing the provided guarantee.

Besides our work on robustness or uncertainty, we also address the problem of 3D shape completion of partially observed point clouds. Specifically, we consider an autonomous driving or robotics setting where vehicles are commonly equipped with LiDAR or depth sensors and obtaining a complete 3D representation of the environment is crucial. However, ground truth shapes that are essential for applying deep learning techniques are extremely difficult to obtain. Thus, we propose a weakly-supervised approach that can be trained on the incomplete point clouds while offering efficient inference.

In summary, this thesis contributes to our understanding of robustness against both input and weight perturbations. To this end, we also develop methods to improve robustness alongside uncertainty estimation for safe deployment of deep learning methods in high-stakes applications. In the particular context of autonomous driving, we also address 3D shape completion of sparse point clouds.

Download & Citing

The thesis is available on the webpage of Saarland University's library:

Table for contents:

If you are interested in the LaTeX templates used for my thesis and defense, you can find them on GitHub:

Thesis Template Defense Template

SEARCH THEBLOG

• 2024 —
• 2023 —
• 2022 —
• 2021 —
• 2020 —
• 2019 —
• 2018 —
• 2017 —
• 2016 —
• 2015 —
• 2014 —
• 2013 —
• ADVERSARIAL MACHINE LEARNING
• ARTIFICIAL INTELLIGENCE
• COMPRESSED SENSING
• COMPUTER GRAPHICS
• COMPUTER SCIENCE
• COMPUTER VISION
• DATA MINING
• DEEP LEARNING
• DNN ACCELERATORS
• GAME THEORY
• IMAGE PROCESSING
• MACHINE LEARNING
• MATHEMATICS
• MEDIA COVERAGE
• MEDICAL IMAGE PROCESSING
• NATURAL LANGUAGE PROCESSING
• NUMERICAL ANALYSIS
• OPTIMIZATION
• PUBLICATION
• RASPBERRY PI
• SECURITY AND PRIVACY
• SOCIAL NETWORKS
• SOFTWARE ENGINEERING
• TWITTER BOOTSTRAP
• UNCERTAINTY ESTIMATION
• WEB SECURITY

• sci.aalto.fi
• cs.aalto.fi

Deep Learning

• Publications

Theses done in the research group

Doctoral theses.

J. Luttinen (2015). Bayesian Latent Gaussian Spatio-Temporal Models . PhD thesis, Aalto University, Espoo, Finland.

K. Cho (2014). Foundations and Advances in Deep Learning . PhD thesis, Aalto University, Espoo, Finland.

M. Harva (2008). Algorithms for Approximate Bayesian Inference with Applications to Astronomical Data Analysis . PhD thesis, Helsinki University of Technology, Espoo, Finland.

T. Raiko (2006). Bayesian Inference in Nonlinear and Relational Latent Variable Models . PhD thesis, Helsinki University of Technology, Espoo, Finland.

A. Ilin (2006). Advanced Source Separation Methods with Applications to Spatio-Temporal Datasets . PhD thesis, Helsinki University of Technology, Espoo, Finland.

A. Honkela (2005). Advances in Variational Bayesian Nonlinear Blind Source Separation . PhD thesis, Helsinki University of Technology, Espoo, Finland.

Master's Theses

R. Boney (2018). Fast Adaptation of Neural Networks . Master's thesis, Aalto University, Helsinki, Finland.

G. van den Broeke (2016). What auto-encoders could learn from brains - Generation as feedback in unsupervised deep learning and inference . Master's thesis, Aalto University, Helsinki, Finland.

M. Perello Nieto (2015). Merging chrominance and luminance in an early, medium and late fusion using Convolutional Neural Networks . Master's thesis, Aalto University, Helsinki, Finland.

P. Noeva (2012). Sampling Methods for Missing Value Reconstruction. Master's thesis, Aalto University, Helsinki, Finland.

T. Hao (2012). Gated Boltzmann Machine in Texture Modelling . Master's thesis, Aalto University, Helsinki, Finland.

K. Cho (2011). Improved Learning Algorithms for Restricted Boltzmann Machines . Master's thesis, Aalto University, Helsinki, Finland.

N. Korsakova (2010). Feature extraction and dynamical modelling of spatio-temporal data. Master's thesis, Aalto University, Finland.

M. Tornio (2010). Natural Gradient for Variational Bayesian Learning . Master's thesis, Aalto University, Helsinki, Finland.

J. Luttinen (2009). Gaussian-process factor analysis for modeling spatio-temporal data . Master's thesis, Helsinki University of Technology, Espoo, Finland.

M. Harva (2004). Hierarchical Variance Models of Image Sequences . Master's thesis, Helsinki University of Technology, Espoo, Finland.

T. Raiko (2001). Hierarchical Nonlinear Factor Analysis . Master's thesis, Helsinki University of Technology, Espoo, Finland.

A. Honkela (2001). Nonlinear Switching State-Space Models . Master's thesis, Helsinki University of Technology, Espoo, Finland.

Latest update 2018-10-10 15:31 EEST     Maintained by Alexander Ilin

File(s) under embargo

Reason: Parts of the thesis are not published yet.

until file(s) become available

DISCOVERY OF NOVEL DISEASE BIOMARKERS AND THERAPEUTICS USING MACHINE LEARNING MODELS

In the fields of computational biology and bioinformatics, the identification of novel disease biomarkers and therapeutic strategies, especially for cancer diseases, remains a crucial challenge. The advancement in computer science, particularly machine learning techniques, has greatly empowered the study of computational biology and bioinformatics for their unprecedented prediction power. This thesis explores how to utilize classic and advanced machine learning models to predict prognostic pathways, biomarkers, and therapeutics associated with cancers.

Firstly, this thesis presents a comprehensive overview of computational biology and bioinformatics, covering past milestones to current groundbreaking advancements, providing context for the research. A centerpiece of this thesis is the introduction of the Pathway Ensemble Tool and Benchmark, an original methodology designated for the unbiased discovery of cancer-related pathways and biomarkers. This toolset not only enhances the identification of crucial prognostic components distinguishing clinical outcomes in cancer patients but also guides the development of targeted drug treatments based on these signatures. Inspired by Benchmark, we extended the methodology to single-cell technologies and proposed scBenchmark and PathPCA, which provides insights into the potential and limitations of how novel techniques can benefit biomarker and therapeutic discovery. Next, the research progresses to the development of DREx, a deep learning model trained on large-scale transcriptome data, for predicting gene expression responses to drug treatments across multiple cell lines. DREx highlights the potential of advanced machine learning models in drug repurposing using a genomics-centric approach, which could significantly enhance the efficiency of initial drug selection.

The thesis concludes by summarizing these findings and highlighting their importance in advancing cancer-related biomarkers and drug discovery. Various computational predictions in this work have already been experimentally validated, showcasing the real-life impact of these methodologies. By integrating machine learning models with computational biology and bioinformatics, this research pioneers new standards for novel biomarker and therapeutics discovery.

Degree Type

• Doctor of Philosophy
• Computer Science

Campus location

• West Lafayette

Advisor/Supervisor/Committee Chair

Additional committee member 2, additional committee member 3, additional committee member 4, usage metrics.

• Other information and computing sciences not elsewhere classified

Help | Advanced Search

Astrophysics > High Energy Astrophysical Phenomena

Title: population synthesis of galactic pulsars with machine learning.

Abstract: This thesis work represents the first efforts to combine population synthesis studies of the Galactic isolated neutron stars with deep-learning techniques with the aim of better understanding neutron-star birth properties and evolution. In particular, we develop a flexible population-synthesis framework to model the dynamical and magneto-rotational evolution of neutron stars, their emission in radio and their detection with radio telescopes. We first study the feasibility of using deep neural networks to infer the dynamical properties at birth and then explore a simulation-based inference approach to predict the birth magnetic-field and spin-period distributions and the late-time magnetic-field decay for the observed radio pulsar population. Our results for the birth magneto-rotational properties agree with the findings of previous works while we constrain the late-time evolution of the magnetic field in neutron stars for the first time. Moreover, this thesis also studies possible scenarios to explain the puzzling nature of recently discovered periodic radio sources with very long periods of the order of thousands of seconds. In particular, by assuming a neutron-star origin, we study the spin-period evolution of a newborn neutron star interacting with a supernova fallback disk and find that the combination of strong, magnetar-like magnetic fields and moderate accretion rates can lead to very large spin periods on timescales of ten thousands of years. Moreover, we perform population synthesis studies to assess the possibility for these sources to be either neutron stars or magnetic white dwarfs emitting coherently through magnetic dipolar losses. These discoveries have opened up a new perspective on the neutron-star population and have started to question our current understanding of how coherent radio emission is produced in pulsar magnetospheres.

Submission history

Access paper:.

• HTML (experimental)
• Other Formats

References & Citations

• INSPIRE HEP
• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

Department of Computer Science

• Department of Computer Science DIKU
• Event Calendar 2024

PhD defence by Lei Li

Follow this link to participate on Zoom

Contributions to deep learning for computer vision applied to environmental remote sensing and human face and pose analysis

This thesis presents research in deep learning for computer vision with applications to remote sensing data and human face and pose analysis. The focus is on 3D data, and various input modalities are considered: images, 3D point clouds, and natural language.

The first part of the thesis considers remote sensing of the environment. The first contribution in this domain is the use of Chain-of-Thought language prompting to enhance semantic image segmentation accuracy, particularly in challenging scenarios like flood disasters. This approach fuses visual and linguistic elements.

The second study considers aligning and fusing diverse modalities for the segmentation of buildings in satellite imagery.

The third study considers the prediction of aboveground forest biomass based on 3D point clouds from airborne LiDAR, for example for measuring carbon sequestration. We suggest replacing the current analysis of handcrafted statistical features derived from the point clouds by applying point cloud neural networks for regression directly to the 3D data. Then edgeaware learning for 3D point clouds is proposed, which addresses the challenges of noise in point cloud data by focusing on edge features to improve classification and segmentation.

The second part of the thesis focuses on the analysis of data from humans. The first work presented involves systems capable of real-time face segmentation, which includes accurate face detection, alignment, and parsing. These systems leverage 3D facial features and can handle occlusions and diverse facial expressions. The final study addresses human pose estimation, which finds extensive application in fields such as augmented reality/virtual reality (AR/VR), live broadcasting, and interactive media. It enhances the user experience by providing more realistic and responsive interactions. The proposed method leverages zero-shot learning algorithms to accurately capture and analyze human movements and poses with diffusion generation methods in uncontrolled environments.

We propose various methods for a range of applied computer vision tasks, utilizing different data modalities. Our research encompasses several scenarios, unified by a core challenge: effectively employing deep learning networks for the analysis of varied modalities.

Supervisors

Principal Supervisor Christian Igel

Assessment Committee

Professor Kim Steenstrup Pedersen, Computer Science Professor Yifang Ban, KTH, Sweden Professor Daniel Sonntag, Oldenburg University, Germany

For an electronic copy of the thesis, please visit  the PhD Programme page . 

Time: 3 May 2024, 13:00-16:00

Place: Zoom

Organizer: Department of Computer Science