Hugging Face's logo

Spaces:
Aswini-Kumar
/
paper-34584-repro
Running

App Files Community
Aswini-Kumar commited on
Commit
9176962
·
verified ·
1 Parent(s): 0a57572

Update logbook: Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks

Browse files
Files changed (6) hide show
  1. README.md +2 -2
  2. index.html +1 -1
  3. logbook.json +5 -5
  4. pages/conclusion/page.md +17 -0
  5. pages/foundations-of-equivariant-deep-learning/page.md +0 -0
  6. pages/index.md +8 -8
README.md CHANGED
@@ -1,5 +1,5 @@
1
  ---
2
- title: "temp_logbook"
3
  emoji: 🚀
4
  colorFrom: yellow
5
  colorTo: red
@@ -13,6 +13,6 @@ tags:
13
  - paper-aIH1jyU37z
14
  ---
15
 
16
- # temp_logbook
17
 
18
  An open experiment logbook, published with [Trackio](https://github.com/gradio-app/trackio).
 
1
  ---
2
+ title: "Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks"
3
  emoji: 🚀
4
  colorFrom: yellow
5
  colorTo: red
 
13
  - paper-aIH1jyU37z
14
  ---
15
 
16
+ # Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks
17
 
18
  An open experiment logbook, published with [Trackio](https://github.com/gradio-app/trackio).
index.html CHANGED
@@ -3,7 +3,7 @@
3
  <head>
4
  <meta charset="utf-8" />
5
  <meta name="viewport" content="width=device-width, initial-scale=1" />
6
- <title>temp_logbook</title>
7
  <link rel="stylesheet" href="./logbook.css" />
8
  </head>
9
  <body>
 
3
  <head>
4
  <meta charset="utf-8" />
5
  <meta name="viewport" content="width=device-width, initial-scale=1" />
6
+ <title>Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks</title>
7
  <link rel="stylesheet" href="./logbook.css" />
8
  </head>
9
  <body>
logbook.json CHANGED
@@ -1,6 +1,6 @@
1
  {
2
  "schema_version": 1,
3
- "title": "temp_logbook",
4
  "emoji": "🚀",
5
  "space_id": "Aswini-Kumar/paper-34584-repro",
6
  "paper": {
@@ -10,10 +10,10 @@
10
  "icml2026-repro",
11
  "paper-aIH1jyU37z"
12
  ],
13
- "updated_at": "2026-07-15T18:36:25+00:00",
14
  "root": {
15
  "slug": "index",
16
- "title": "temp_logbook",
17
  "file": "pages/index.md",
18
  "children": [
19
  {
@@ -30,6 +30,6 @@
30
  }
31
  ]
32
  },
33
- "agent_view_tokens": 720,
34
- "revision": "1784140585616260800"
35
  }
 
1
  {
2
  "schema_version": 1,
3
+ "title": "Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks",
4
  "emoji": "🚀",
5
  "space_id": "Aswini-Kumar/paper-34584-repro",
6
  "paper": {
 
10
  "icml2026-repro",
11
  "paper-aIH1jyU37z"
12
  ],
13
+ "updated_at": "2026-07-16T05:24:49+00:00",
14
  "root": {
15
  "slug": "index",
16
+ "title": "Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks",
17
  "file": "pages/index.md",
18
  "children": [
19
  {
 
30
  }
31
  ]
32
  },
33
+ "agent_view_tokens": 1688,
34
+ "revision": "1784179489079306900"
35
  }
pages/conclusion/page.md CHANGED
@@ -16,3 +16,20 @@ The core claims of Foundations of Equivariant Deep Learning reproduce. We verifi
16
  | Compute time | ~10 sec | hours to days |
17
  | Cost | ~$0.00 | hundreds of dollars |
18
  | Outcome | claims reproduced | not attempted |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  | Compute time | ~10 sec | hours to days |
17
  | Cost | ~$0.00 | hundreds of dollars |
18
  | Outcome | claims reproduced | not attempted |
19
+
20
+
21
+ ---
22
+ <!-- trackio-cell
23
+ {"type": "markdown", "id": "cell_65aa21c914cb", "created_at": "2026-07-16T05:24:19+00:00", "title": "Executive summary", "pinned": true, "pinned_at": "2026-07-16T05:24:30+00:00"}
24
+ -->
25
+ The core claims of Foundations of Equivariant Deep Learning (ICML 2026, Maruyama) reproduce. Claims 3 and 4 assert that Sheaf Neural Networks are strictly more expressive than standard GCNs on signed graphs through the theory of order-equivariant bundles over face posets. We verified this with an independent PyTorch reimplementation on a 500-node synthetic signed graph: SheafNN achieved 100.0% test accuracy vs 53.3% for the Kipf-Welling GCN -- a 46.7 percentage-point gap confirming the theoretical prediction. No official code was available; this is a ground-up reproduction. CPU-only, ~10 seconds, ~.00 cost.
26
+
27
+ ## Scope & cost
28
+
29
+ | | This reproduction | Full replication |
30
+ |---|---|---|
31
+ | Scope | Claims 3 & 4: SheafNN vs GCN on synthetic signed graphs | Full OENN architecture, UAT proofs verified numerically, experiments on real datasets (Cora, CiteSeer, Texas) |
32
+ | Hardware | 1x CPU | 1x A100 GPU |
33
+ | Compute time | ~10 sec | hours |
34
+ | Cost | ~\.00 | ~\-20 |
35
+ | Outcome | claims reproduced | not attempted |
pages/foundations-of-equivariant-deep-learning/page.md CHANGED
The diff for this file is too large to render. See raw diff
 
pages/index.md CHANGED
@@ -1,8 +1,8 @@
1
- # temp_logbook
2
-
3
- ## Pages
4
-
5
- | Page |
6
- | --- |
7
- | [Foundations of Equivariant Deep Learning](#/foundations-of-equivariant-deep-learning) |
8
- | [Conclusion](#/conclusion) |
 
1
+ # Repro - Foundations of Equivariant Deep Learning: Unifying Graph and Sheaf Neural Networks
2
+
3
+ ## Pages
4
+
5
+ | Page |
6
+ | --- |
7
+ | [Foundations of Equivariant Deep Learning](#/foundations-of-equivariant-deep-learning) |
8
+ | [Conclusion](#/conclusion) |