Research:
The 21st century is revolutionizing data analysis and information processing (cloud, big data, privacy, HPC, data science to cite a few)
and open brand new horizons for adding novel "values" to data or extracting "information" thereof, with both industrial and societal impacts.
I am addressing those topics under the umbrella of Computational Information Geometry (CIG)
with applications to machine learning, computational statistics
and imaging (including computer vision), mainly.
But:
- what is information?
- what is computation? and
- what is geometry?
Furthermore, how do those concepts interact with each others?
These are fascinating and tantalizing questions since it is natural to ask oneself whether current data analysis methods are "optimal" or "biased" (reliable?)
and what kind of information
can be extracted from data?
The geometric framework allows us to perform intrinsic analysis that are not biased by the ad hoc data representations and
to analyze the data landscape:
We are investigating the geometric structures of data spaces for pure analytics.
I am writing software (C++/Java/Processing/R/Python) and applications to demonstrate those capabilities.
My contributions and interests are:
- Machine learning (classification, clustering)
- Computational statistics
- Computer vision (image segmentation, image retrieval systems, etc.)
- Computational photography (image-based rendering, surround panoramic video, free 3D viewpoint, etc.)
- Human Computer interaction (image editing, login interfaces, etc.)
- Computational geometry
- Information geometry
- Geometric combinatorial optimization (antenna network layout)
- etc.
- Upcoming organizing events:
- Past events:
- ICMS workshop on Computational Information Geometry for Image and Signal Processing (CIGISP), 21-25 September, Edinburgh, UK, 2015.
- Geometric Science of Information (GSI), Oct 28-30 (Wed-Fri), Ecole Polytechnique, France, 2015.
- What is new?
- [publications chronological | reverse]
[slides]
[video]
[DBLP]
[software]
- Preprints:
- Computational Information Geometry for Machine Learning (MLSS 2015, Sydney)
- Some demos with source codes:
- Papers:
- Loss factorization, weakly supervised learning and label noise robustness, ICML 2016
- k-variates++: more pluses in the k-means++, ICML 2016
- Which Geometry for Clustering Copulas? (Gimli:Geometry in Machine Learning workshop, ICML 2016)
- Guaranteed bounds on the Kullback-Leibler divergence of univariate mixtures using piecewise log-sum-exp inequalities, preprint, 2016
(arxiv 1606.05850), IEEE Signal Processing Letters,
pdf xplore
- Learning on High-dimensional Neuromanifolds with Relative Natural Gradients
(Gimli:Geometry in Machine Learning workshop, ICML 2016) (arxiv 1606.06069")
-
Clustering Financial Time Series: How Long is Enough? (IJCAI'16)
- Optimal transport vs.
Fisher-Rao distance between copulas for clustering multivariate time series (IEEE SSP'16)
- Classification with mixtures of curved Mahalanobis Metrics
(IEEE ICIP'16)
- SSSC-AM:
A unified framework for video co-segmentation by structured sparse subspace clustering with appearance and motion features (IEEE ICIP'16)
- Quantifying the Invariance and Robustness of Permutation-Based Indexing Schemes (SISAP'16, LNCS)
-
Patch Matching with Polynomial Exponential Families and Projective Divergences (SISAP'16, LNCS)
- Optimal Copula Transport for Clustering Multivariate Time Series, IEEE ICASSP 2016.
- Comix: Joint Estimation and Lightspeed Comparison of Mixture Models, IEEE ICASSP 2016.
- Image and information, preprint, 2016
- Bridging weak supervision and privacy aware learning via sufficient statistics
NIPS 2015, workshop on Learning and privacy with incomplete data and weak supervision
- A proposal of a methodological framework with experimental guidelines to investigate clustering stability on financial time series,
14th International Conference on Machine Learning and Applications (IEEE ICMLA'15)
- HCMapper: An interactive visualization tool to compare partition-based flat
clustering extracted from pairs of dendrograms. CoRR abs/1507.08137 (2015)
- TS-GNPR Clustering Random Walk Time Series
slides
video
- Bag-of-components an online algorithm for batch learning of mixture models
slides
video
- Approximating Covering and Minimum Enclosing Balls in Hyperbolic Geometry
slides
video
- Online k-MLE for mixture modeling with exponential families
slides
video
- Estimation jointe et en ligne de modèles de mélanges avec les
co-mélanges et les sacs de composantes, GRETSI 2015
- Comment partitionner automatiquement des marches aléatoires ?
Avec application à la finance quantitative , GRETSI 2015
- TOTAL JENSEN DIVERGENCES: DEFINITION, PROPERTIES AND CLUSTERING (ICASSP'2015)
- On learning statistical mixtures maximizing the
complete likelihood, MaxEnt (video), AIP Proceedings (2014)
[Paper]
[BibTeX]
[Summary]
[Slides]
[Video]
- On Clustering Histograms with k-Means by Using Mixed \alpha-Divergences.
MDPI Entropy 16(6): 3273-3301 (2014)
[Paper]
[BibTeX]
[Summary]
[Slides]
[code Java/Processing]
- Generalized Bhattacharyya and Chernoff upper bounds on Bayes error using quasi-arithmetic means. Elsevier Pattern Recognition Letters 42: 25-34 (2014)
[paper]
[BibTeX]
[Summary]
[Slides]
[Code Java]
- On the Chi Square and Higher-Order Chi Distances for Approximating f-Divergences. IEEE Signal Process. Lett. 21(1): 10-13 (2014)
[Paper]
[BibTeX]
[Summary]
[Slides]
[Code Java]
[code Java]
- Optimal Interval Clustering: Application to Bregman Clustering and Statistical Mixture Learning. IEEE Signal Process. Lett. 21(10): 1289-1292 (2014)
[BibTeX]
[Summary]
[Slides]
[Poster]
[code Java]
- Visualizing hyperbolic Voronoi diagrams. ACM Symposium on Computational Geometry (2014)
[BibTeX]
[Paper]
[Video]
[Web]
- On approximating the Riemannian 1-center. Comput. Geom. 46(1): 93-104 (2013)
[BibTeX]
[Paper]
[Slides]
[code Java/Processing]
- An Information-Geometric Characterization of Chernoff Information. IEEE Signal Process. Lett. 20(3): 269-272 (2013)
[BibTeX]
[Paper]
[Slides]
[code Java]
- Jeffreys Centroids: A Closed-Form Expression for Positive Histograms and a Guaranteed Tight Approximation for Frequency Histograms. IEEE Signal Process. Lett. 20(7): 657-660 (2013)
[BibTeX]
[Paper]
[Slides]
[code Java/R]
- Consensus Region Merging for Image Segmentation. ACPR 2013: 325-329 (2013)
[BibTeX]
[Paper]
[code Java]
- Perspective click-and-drag area selections in pictures. MVA 2013: 29-32
[BibTeX]
[Paper]
[Slides]
[Video]
[code Java/Processing]
Best Practical Paper Award
- Logging safely in public spaces using color PINs. CoRR abs/1304.6499 (2013)
[Paper]
[Video]
[code Java/Processing]
- more from DBLP
- Overview:
- Editors:
Journal of Mathematical Imaging and Vision (JMIV)
MDPI Entropy (JMIV)
GPU:
- Teaching:
© 2014 Frank Nielsen, all rights reserved. former home page
November 2014, gs bib 09/2016
