Skip to main content
Cornell University
We gratefully acknowledge support from the Simons Foundation, member institutions, and all contributors. Donate
arxiv logo > cs > arXiv:1901.08556

Help | Advanced Search

arXiv logo
Cornell University Logo

quick links

  • Login
  • Help Pages
  • About

Computer Science > Machine Learning

arXiv:1901.08556 (cs)
[Submitted on 20 Jan 2019]

Title:Visualized Insights into the Optimization Landscape of Fully Convolutional Networks

Authors:Jianjie Lu, Kai-yu Tong
View a PDF of the paper titled Visualized Insights into the Optimization Landscape of Fully Convolutional Networks, by Jianjie Lu and 1 other authors
View PDF
Abstract:Many image processing tasks involve image-to-image mapping, which can be addressed well by fully convolutional networks (FCN) without any heavy preprocessing. Although empirically designing and training FCNs can achieve satisfactory results, reasons for the improvement in performance are slightly ambiguous. Our study is to make progress in understanding their generalization abilities through visualizing the optimization landscapes. The visualization of objective functions is obtained by choosing a solution and projecting its vicinity onto a 3D space. We compare three FCN-based networks (two existing models and a new proposed in this paper for comparison) on multiple datasets. It has been observed in practice that the connections from the pre-pooled feature maps to the post-upsampled can achieve better results. We investigate the cause and provide experiments to shows that the skip-layer connections in FCN can promote flat optimization landscape, which is well known to generalize better. Additionally, we explore the relationship between the models generalization ability and loss surface under different batch sizes. Results show that large-batch training makes the model converge to sharp minimizers with chaotic vicinities while small-batch method leads the model to flat minimizers with smooth and nearly convex regions. Our work may contribute to insights and analysis for designing and training FCNs.
Comments: In AAAI-19 Workshop on Network Interpretability for Deep Learning
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:1901.08556 [cs.LG]
  (or arXiv:1901.08556v1 [cs.LG] for this version)
  https://doi.org/10.48550/arXiv.1901.08556
arXiv-issued DOI via DataCite

Submission history

From: Quanshi Zhang [view email]
[v1] Sun, 20 Jan 2019 18:49:34 UTC (3,365 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled Visualized Insights into the Optimization Landscape of Fully Convolutional Networks, by Jianjie Lu and 1 other authors
  • View PDF
  • TeX Source
  • Other Formats
view license
Current browse context:
cs.LG
< prev   |   next >
new | recent | 2019-01
Change to browse by:
cs
stat
stat.ML

References & Citations

  • NASA ADS
  • Google Scholar
  • Semantic Scholar

DBLP - CS Bibliography

listing | bibtex
Jianjie Lu
Kai-yu Tong
export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo

Bibliographic and Citation Tools

Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)

Code, Data and Media Associated with this Article

alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)

Demos

Replicate (What is Replicate?)
Hugging Face Spaces (What is Spaces?)
TXYZ.AI (What is TXYZ.AI?)

Recommenders and Search Tools

Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)
IArxiv Recommender (What is IArxiv?)
  • Author
  • Venue
  • Institution
  • Topic

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.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)
  • About
  • Help
  • contact arXivClick here to contact arXiv Contact
  • subscribe to arXiv mailingsClick here to subscribe Subscribe
  • Copyright
  • Privacy Policy
  • Web Accessibility Assistance
  • arXiv Operational Status
    Get status notifications via email or slack