Update README.md
Browse files
README.md
CHANGED
|
@@ -17,7 +17,37 @@ tags:
|
|
| 17 |
pretty_name: Multimodal Unlearning Evaluation Benchmark
|
| 18 |
size_categories:
|
| 19 |
- n<1K
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
---
|
|
|
|
| 21 |
# 🧠 Multimodal Unlearning Evaluation Benchmark
|
| 22 |
|
| 23 |
## 📌 Overview
|
|
@@ -32,11 +62,11 @@ It supports reproducibility of results in:
|
|
| 32 |
|
| 33 |
| File | Description |
|
| 34 |
|------|------------|
|
| 35 |
-
| 📄 `multimodal_results.json` | Results on VQA benchmarks (MLLMU, UnLOK, MMUBench) |
|
| 36 |
| 📄 `unimodal_results.json` | CIFAR-10 baseline results |
|
| 37 |
| ⚖️ `uqs_weights.json` | Learned weights for Unified Quality Score (UQS) |
|
| 38 |
| 🏆 `ranking_table.json` | Method rankings across metrics |
|
| 39 |
-
| 📈 `analysis_results.json` | Correlation
|
| 40 |
| 🔍 `kr_pilot_results.json` | Knowledge Recoverability (KR) pilot results |
|
| 41 |
| 🤖 `blip2_minimal_summary.json` | Cross-architecture validation (BLIP-2) |
|
| 42 |
|
|
|
|
| 17 |
pretty_name: Multimodal Unlearning Evaluation Benchmark
|
| 18 |
size_categories:
|
| 19 |
- n<1K
|
| 20 |
+
configs:
|
| 21 |
+
- config_name: multimodal
|
| 22 |
+
data_files:
|
| 23 |
+
- split: train
|
| 24 |
+
path: multimodal_results.json
|
| 25 |
+
- config_name: unimodal
|
| 26 |
+
data_files:
|
| 27 |
+
- split: train
|
| 28 |
+
path: unimodal_results.json
|
| 29 |
+
- config_name: uqs_weights
|
| 30 |
+
data_files:
|
| 31 |
+
- split: train
|
| 32 |
+
path: uqs_weights.json
|
| 33 |
+
- config_name: ranking
|
| 34 |
+
data_files:
|
| 35 |
+
- split: train
|
| 36 |
+
path: ranking_table.json
|
| 37 |
+
- config_name: analysis
|
| 38 |
+
data_files:
|
| 39 |
+
- split: train
|
| 40 |
+
path: analysis_results.json
|
| 41 |
+
- config_name: kr_pilot
|
| 42 |
+
data_files:
|
| 43 |
+
- split: train
|
| 44 |
+
path: kr_pilot_results.json
|
| 45 |
+
- config_name: blip2
|
| 46 |
+
data_files:
|
| 47 |
+
- split: train
|
| 48 |
+
path: blip2_minimal_summary.json
|
| 49 |
---
|
| 50 |
+
|
| 51 |
# 🧠 Multimodal Unlearning Evaluation Benchmark
|
| 52 |
|
| 53 |
## 📌 Overview
|
|
|
|
| 62 |
|
| 63 |
| File | Description |
|
| 64 |
|------|------------|
|
| 65 |
+
| 📄 `multimodal_results.json` | Results on VQA benchmarks (MLLMU-Bench, UnLOK-VQA, MMUBench) |
|
| 66 |
| 📄 `unimodal_results.json` | CIFAR-10 baseline results |
|
| 67 |
| ⚖️ `uqs_weights.json` | Learned weights for Unified Quality Score (UQS) |
|
| 68 |
| 🏆 `ranking_table.json` | Method rankings across metrics |
|
| 69 |
+
| 📈 `analysis_results.json` | Correlation and disagreement analysis |
|
| 70 |
| 🔍 `kr_pilot_results.json` | Knowledge Recoverability (KR) pilot results |
|
| 71 |
| 🤖 `blip2_minimal_summary.json` | Cross-architecture validation (BLIP-2) |
|
| 72 |
|