Thumb ticker lg profile  1

Id: 999592

User: 448

Login: lballes

Image: /uploads/is_web_engine/person/image/999592/profile__1_.jpg

Title:

Firstname: Lukas

Lastname: Balles

Room: S2.018

Street: Max-Planck-Ring 4

Location: Tübingen

Ldap location:

Postcode: 72076

Email: lukas.balles@tuebingen.mpg.de

Mobile:

Telephone: +49 7071 601 1714

Fax: +49 7071 601 552

Start date:

End date:

Small description:

Description: <p>This page is no longer being maintained. Please refer to my webpage&nbsp;<a href="https://lballes.github.io">https://lballes.github.io</a>.</p> <h3>About Me</h3> <p>I am a Ph.D. student in the <a href="http://pn.is.tuebingen.mpg.de">Probabilistic Numerics Group</a>, where I work on optimization methods for machine learning. Optimization algorithms are the workhorse of contemporary machine learning - it&#39;s where the numbers are crunched! Intriguingly, numerical optimizers themselves are compact little &quot;learning machines&quot;: they make decisions (where to evaluate next, how many and which data points to use) based on observations (function values and gradients, typically corrupted by noise due to mini-batch subsampling). My goal is to design smarter optimizers!</p> <p>Currently, my work centers around estimating the stochastic gradient variance, a measure for how &quot;noisy&quot; stochastic gradients are. I believe that using this quantity to make optimizers <em>aware</em> of the stochasticity of their evaluations can help improve various aspects of stochastic optimization algorithms. For example, gradient variance estimates can be used to adaptively choose &quot;good&quot; batch sizes when performing stochastic gradient descent \cite{balles2017coupling}. Element-wise variance estimates can also be used to manipulate&nbsp;the search direction itself by &quot;damping&quot; coordinate directions&nbsp;with low signal-to-noise ratio \cite{balles2018dissecting}.</p> <p>Prior to joining the Max Planck Institute as a Ph.D. student, I studied Mathematics (B.Sc.) and Scientific Computing (M.Sc.) at Heidelberg University and spent some time as a visiting student at Tsinghua University in Beijing.</p>

Keywords:

Meta description:

Meta keywords:

Footnote:

Link: https://lballes.github.io

Facebook url:

Linkedin url: http://linkedin.com/in/lballes

Xing url:

Twitter url: http://twitter.com/lukas_balles

Googlescholar url: https://scholar.google.de/citations?user=2lq9JQIAAAAJ&hl=de

Github url: http://github.com/lballes

Skype name:

Position: #<IsWebEngine::Position::ActiveRecord_Associations_CollectionProxy:0x007f8de5f30870>

Advisor:

Student description:

Alumni: true

Hidden: false

Activated: true

Show linked employee: true

Show publications: true

Publication show images: true

Show research projects: false

Slug: lballes

Edit | Back