Skip to content

Commit 38477dc

Browse files
committed
Merge branch 'source' of github.com:UTAustin-SwarmLab/UTAustin-SwarmLab.github.io into source
2 parents ac35e20 + 3f8ae03 commit 38477dc

File tree

11 files changed

+151
-30
lines changed

11 files changed

+151
-30
lines changed

.github/workflows/jekyll.yml

Lines changed: 38 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -1,47 +1,65 @@
1+
# This workflow uses actions that are not certified by GitHub.
2+
# They are provided by a third-party and are governed by
3+
# separate terms of service, privacy policy, and support
4+
# documentation.
5+
6+
# Sample workflow for building and deploying a Jekyll site to GitHub Pages
17
name: Deploy Jekyll site to Pages
28

39
on:
10+
# Runs on pushes targeting the default branch
411
push:
512
branches: ["source"]
613

14+
# Allows you to run this workflow manually from the Actions tab
15+
workflow_dispatch:
16+
17+
# Sets permissions of the GITHUB_TOKEN to allow deployment to GitHub Pages
718
permissions:
819
contents: read
920
pages: write
1021
id-token: write
1122

23+
# Allow only one concurrent deployment, skipping runs queued between the run in-progress and latest queued.
24+
# However, do NOT cancel in-progress runs as we want to allow these production deployments to complete.
1225
concurrency:
1326
group: "pages"
14-
cancel-in-progress: true
27+
cancel-in-progress: false
1528

1629
jobs:
30+
# Build job
1731
build:
1832
runs-on: ubuntu-latest
1933
steps:
2034
- name: Checkout
2135
uses: actions/checkout@v4
22-
2336
- name: Setup Ruby
24-
uses: ruby/setup-ruby@v1
37+
# https://github.com/ruby/setup-ruby/releases/tag/v1.207.0
38+
uses: ruby/setup-ruby@4a9ddd6f338a97768b8006bf671dfbad383215f4
2539
with:
26-
ruby-version: '3.4.1'
27-
bundler-cache: true
28-
cache-version: 1
29-
40+
ruby-version: '3.2.3' # Not needed with a .ruby-version file
41+
bundler-cache: true # runs 'bundle install' and caches installed gems automatically
42+
cache-version: 0 # Increment this number if you need to re-download cached gems
43+
- name: Setup Pages
44+
id: pages
45+
uses: actions/configure-pages@v5
3046
- name: Build with Jekyll
47+
# Outputs to the './_site' directory by default
3148
run: bundle exec jekyll build --baseurl "${{ steps.pages.outputs.base_path }}"
3249
env:
3350
JEKYLL_ENV: production
51+
- name: Upload artifact
52+
# Automatically uploads an artifact from the './_site' directory by default
53+
uses: actions/upload-pages-artifact@v3
3454

35-
# deploy:
36-
# needs: build
37-
# runs-on: ubuntu-latest
38-
# permissions:
39-
# pages: write
40-
# id-token: write
41-
# environment:
42-
# name: github-pages
43-
# url: ${{ steps.deployment.outputs.page_url }}
44-
# steps:
45-
# - name: Deploy to GitHub Pages
46-
# id: deployment
47-
# uses: actions/deploy-pages@v4
55+
# Deployment job
56+
deploy:
57+
environment:
58+
name: github-pages
59+
url: ${{ steps.deployment.outputs.page_url }}
60+
runs-on: ubuntu-latest
61+
needs: build
62+
steps:
63+
- name: Deploy to GitHub Pages
64+
id: deployment
65+
uses: actions/deploy-pages@v4

Gemfile

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
# frozen_string_literal: true
22

33
source "https://rubygems.org"
4-
ruby ">= 3.2.3"
4+
ruby ">= 3.1.3"
55

66
# gem "rails"
77

_bibliography/references.bib

Lines changed: 29 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,34 @@
22
---
33
references
44
==========
5+
@article{baser2025fairsynergy,
6+
title={FairSynergy: Fair Resource Allocation for Fleet Intelligence},
7+
author={Baser, Oguzhan and Kale, Kaan and Li, Po-han and Chinchali, Sandeep},
8+
journal={IEEE Global Communications Conference (GLOBECOM)},
9+
year={2025},
10+
}
11+
12+
@article{baser2025phonemefake,
13+
title={PhonemeFake: Redefining Deepfake Realism with Language-Driven Segmental Manipulation and Adaptive Bilevel Detection},
14+
author={Baser, Oguzhan and Tanriverdi, Ahmet Ege and Vishwanath, Sriram and Chinchali, Sandeep},
15+
journal={Proceedings of Interspeech},
16+
year={2025},
17+
}
18+
19+
@article{baser2025wavshape,
20+
title={WavShape: Information-Theoretic Speech Representation Learning for Fair and Privacy-Aware Audio Processing},
21+
author={Baser, Oguzhan and Tanriverdi, Ahmet Ege and Kale, Kaan and Chinchali, Sandeep and Vishwanath, Sriram},
22+
journal={Proceedings of Interspeech},
23+
year={2025},
24+
}
25+
26+
@article{akcin2025dual,
27+
title={Distributed Upload and Active Labeling for Resource-Constrained Fleet Learning},
28+
author={Akcin, Oguzhan and Goel, Harsh and Zhao, Ruihan and Sandeep P. Chinchali},
29+
journal={9th Annual Conference on Robot Learning},
30+
year={2025},
31+
url={https://openreview.net/forum?id=M1e2PEMLp2}
32+
}
533

634
@article{Omama2024exploiting,
735
title={Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval},
@@ -35,7 +63,7 @@ @article{akcin2024fleet
3563
@article{baser2024securespectra,
3664
title={SecureSpectra: Safeguarding Digital Identity from Deep Fake Threats via Intelligent Signatures},
3765
author={Baser, Oguzhan and Kale, Kaan and Chinchali, Sandeep},
38-
journal={INTERSPEECH},
66+
journal={Proceedings of Interspeech},
3967
year={2024}
4068
}
4169

_data/news.yml

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,12 @@
1-
- date: 2025-24-01
1+
- date: 2025-08-01
22
details: >-
3-
Po-han's and Omama's papers <a href="https://openreview.net/forum?id=6Mg7pjG7Sw"> CSA: Data-efficient Mapping of Unimodal Features to Multimodal Features </a> and <a href="https://openreview.net/forum?id=d0tlL0ZWlu"> Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval </a> were accepted to ICLR 2025!
4-
- date: 2024-19-12
3+
Oguzhan B.'s paper <a href="https://openreview.net/forum?id=M1e2PEMLp2"> Fair Resource Allocation for Fleet Intelligence </a> was accepted to GLOBECOM 2025!
4+
- date: 2025-08-01
55
details: >-
6-
Philip's papers <a href="https://arxiv.org/abs/2410.18242"> Human-Agent Coordination in Games under Incomplete Information via Multi-Step Intent </a> was accepted to AAMAS 2025!
7-
- date: 2024-19-09
6+
Oguzhan A.'s paper <a href="https://openreview.net/forum?id=M1e2PEMLp2"> Distributed Upload and Active Labeling for Resource-Constrained Fleet Learning </a> was accepted to CoRL 2025!
7+
- date: 2025-07-15
88
details: >-
9-
Philip's and Oguzhan A.'s' papers <a href="https://philipzrh.com/lane/"> Accelerating Visual Sparse-Reward Learning with Latent Nearest-Demonstration-Guided Exploration </a> and <a href="https://openreview.net/forum?id=9dsBQhoqVr"> Fleet Supervisor Allocation: A Submodular Maximization Approach</a> were accepted to CoRL 2024.
9+
Oguzhan B.'s paper <a href="https://arxiv.org/abs/2506.22789"> Information-Theoretic Speech Representation Learning for Fair and Privacy-Aware Audio Processing </a> was accepted to INTERSPEECH 2025!
10+
- date: 2025-07-15
11+
details: >-
12+
Oguzhan B.'s paper <a href="https://arxiv.org/abs/2506.22783"> Redefining Deepfake Realism with Language-Driven Segmental Manipulation and Adaptive Bilevel Detection </a> was accepted to INTERSPEECH 2025!

_people/oguzhan_akcin.md

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
---
22
name: Oguzhan Akcin
3-
position: gradstudent
3+
position: alumni
44
avatar: oguzhan_akcin.jpg
55
---
66

_posts/2025-04-01-IRDC.md

Lines changed: 71 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,71 @@
1+
---
2+
title: 'Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval'
3+
description: ICLR 2025 paper
4+
categories: blog
5+
---
6+
7+
*By Mohammad Omama*
8+
9+
10+
# Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval
11+
12+
> Proceedings of the International Conference on Learning Representations (ICLR)
13+
14+
15+
`TLDR:` Image Retrieval with Foundation Models: Better, Faster, Distribution-Aware!
16+
17+
[ArXiv](https://arxiv.org/abs/2410.07022)
18+
19+
[Project Website](https://mohdomama.github.io/IRDC-Project-Website/)
20+
21+
## Motivation
22+
23+
Image retrieval is pivotal in many real-world applications, from visual place recognition in robotics to personalized recommendations in e-commerce. However, current state-of-the-art (SOTA) image retrieval methods face two significant problems:
24+
25+
1. **Scalability Issue**: State-of-the-art (SOTA) image retrieval methods train large models separately for each dataset. This is __not scalable__.
26+
27+
2. **Efficiency Issue**: SOTA image retrieval methods use large embeddings, and since retrieval speed is directly proportional to embedding size, this is __not efficient__.
28+
29+
Our research specifically targets these challenges with two crucial questions:
30+
- **Q1 (Scalability)**: Can we enhance the performance of universal off-the-shelf models in an entirely unsupervised way?
31+
- **Q2 (Efficiency)**: Is it possible to design an effective unsupervised dimensionality reduction method that preserves the similarity structure and can adaptively perform well at varying embedding sizes?
32+
33+
## Contributions
34+
35+
To tackle the scalability and efficiency challenges, our work introduces the follwoing novel ideas:
36+
37+
- **Autoencoders with Strong Variance Constraints (AE-SVC)**: Addressing scalability, AE-SVC significantly improves off-the-shelf foundation model embeddings through three rigorously enforced constraints: orthogonality, mean-centering, and unit variance in the latent space. We both empirically demonstrate and mathematically validate that these constraints adjust the distribution of cosine similarity, making embeddings more discriminative.
38+
39+
- **Single Shot Similarity Space Distillation ((SS)<sub>2</sub> D)**: To tackle efficiency, (SS)<sub>2</sub> D provides dimensionality reduction that preserves similarity structures
40+
and further allows embeddings to adaptively scale without retraining. This enables smaller segments of the embedding to retain high retrieval performance, significantly speeding up retrieval.
41+
42+
## Methodology
43+
44+
Our proposed approach follows a two-step pipeline:
45+
46+
1. **AE-SVC** first trains an autoencoder with the constraints mentioned to enhance the embeddings from foundation models.
47+
2. The improved embeddings from AE-SVC are then distilled using **(SS)<sub>2</sub>D**, producing embeddings that are both efficient and adaptive at various sizes.
48+
49+
The training process ensures that the resulting embeddings, even at smaller sizes, preserve similarity relationships, making them highly effective for retrieval tasks.
50+
51+
![Methodology Pipeline](/images/post/IRDC_pipeline.png)
52+
53+
## Impact on Cosine Similarity Distribution
54+
55+
Our AE-SVC method profoundly impacts cosine similarity distributions, significantly reducing their variance.
56+
Lower variance in similarity distributions correlates with improved discriminative power as we mathematically prove in our paper.
57+
Our method shows remarkable benefits, particularly for general-purpose foundation models like DINO, compared to already optimized dataset-specific models such as Cosplace.
58+
59+
![Cosine Similarity Distribution](/images/post/IRDC_theory.png)
60+
61+
## Results
62+
63+
Our experimental validation demonstrates:
64+
65+
- **AE-SVC** consistently surpasses baseline PCA methods across multiple datasets, offering an average of 15.5% improvement in retrieval performance.
66+
- **(SS)<sub>2</sub>D**, building upon AE-SVC, achieves up to a 10% further improvement at smaller embedding sizes, demonstrating superior performance compared to traditional dimensionality reduction methods like VAE and approaches the theoretical upper bound set by SSD.
67+
68+
This advancement represents a significant step towards more practical, scalable, and efficient image retrieval solutions, enhancing both speed and accuracy.
69+
70+
![Retrieval Performance Results](/images/post/IRDC_results.png)
71+

images/post/IRDC_pipeline.png

292 KB
Loading

images/post/IRDC_results.png

164 KB
Loading

images/post/IRDC_theory.png

140 KB
Loading

index.html

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,7 @@ <h2> Recent News </h2>
2424
<u>Happenings of the last four months. Updated {{ 'now' | date: "%m/%d/%y" }}.</u>
2525
<span style="display: block; margin-bottom: 1em"></span>
2626
<div class="news">
27-
{% capture now %}{{'now' | date: "%s" | minus: 5184000}}{% endcapture %}
27+
{% capture now %}{{'now' | date: "%s" | minus: 10368000}}{% endcapture %}
2828
<ul style="list-style-position:outside;padding:20px" >
2929
{% for new in site.data.news %}
3030
{% capture date %}{{new.date | date: '%s' | plus: 0}}{% endcapture %}

0 commit comments

Comments
 (0)