Dataset Viewer
Auto-converted to Parquet Duplicate
dataset
stringclasses
3 values
method
stringclasses
4 values
seed
int64
42
5.51k
FA
float64
0
1
RA
float64
0.02
1
MIA
float64
0
0.54
AD
float64
3.78
188
JS
float64
0
0.42
retrained_distance
float64
0.02
0.37
mllmu_bench
gradient_ascent
42
0
0.02
0.38
121.4375
0.020695
0.054101
mllmu_bench
random_labels
42
0
0.02
0.48
17.359375
0.020589
0.054499
mllmu_bench
finetune_retain
42
0
0.02
0.52
26.28125
0.003063
0.061589
mllmu_bench
salun
42
0
0.02
0.5
66.4375
0.020032
0.365435
mllmu_bench
gradient_ascent
123
0
0.02
0.36
122.9375
0.018601
0.054169
mllmu_bench
random_labels
123
0
0.02
0.48
16.921875
0.013238
0.054522
mllmu_bench
finetune_retain
123
0
0.02
0.52
26.046875
0.002479
0.061353
mllmu_bench
salun
123
0
0.02
0.46
65.25
0.009408
0.365397
mllmu_bench
gradient_ascent
5,508
0
0.02
0.32
114.875
0.024985
0.053953
mllmu_bench
random_labels
5,508
0
0.02
0.48
17.140625
0.01659
0.05451
mllmu_bench
finetune_retain
5,508
0
0.02
0.54
25.65625
0.003054
0.061484
mllmu_bench
salun
5,508
0
0.02
0.48
63.6875
0.025928
0.365671
unlok_vqa
gradient_ascent
42
1
1
0
187.75
0.175358
0.01763
unlok_vqa
random_labels
42
1
1
0
77.625
0.094831
0.020989
unlok_vqa
finetune_retain
42
1
1
0
123.5
0.414358
0.05013
unlok_vqa
salun
42
1
1
0
63.8125
0.096371
0.351211
unlok_vqa
gradient_ascent
123
1
1
0
187.125
0.381793
0.017659
unlok_vqa
random_labels
123
1
1
0
76.6875
0.098597
0.020489
unlok_vqa
finetune_retain
123
1
1
0
123.3125
0.418465
0.049774
unlok_vqa
salun
123
1
1
0
68.125
0.179481
0.351114
unlok_vqa
gradient_ascent
5,508
1
1
0
188.25
0.158545
0.017723
unlok_vqa
random_labels
5,508
1
1
0
76.4375
0.092216
0.020617
unlok_vqa
finetune_retain
5,508
1
1
0
123.5
0.418205
0.049936
unlok_vqa
salun
5,508
1
1
0
67
0.097172
0.351184
mmubench
gradient_ascent
42
0.866667
0.84
0.466667
9.328125
0.000189
0.017604
mmubench
random_labels
42
0.833333
0.86
0.533333
29.921875
0.00257
0.021638
mmubench
finetune_retain
42
0.866667
0.72
0.533333
112.5
0.053203
0.043102
mmubench
salun
42
0.866667
0.82
0.466667
22.71875
0.006401
0.352574
mmubench
gradient_ascent
123
0.866667
0.84
0.466667
3.777344
0.000099
0.017637
mmubench
random_labels
123
0.833333
0.86
0.533333
30.3125
0.002563
0.021635
mmubench
finetune_retain
123
0.833333
0.72
0.466667
110.8125
0.053304
0.04294
mmubench
salun
123
0.866667
0.74
0.466667
21.46875
0.004215
0.352633
mmubench
gradient_ascent
5,508
0.866667
0.84
0.466667
9.929688
0.000208
0.017593
mmubench
random_labels
5,508
0.833333
0.86
0.533333
30.203125
0.002564
0.021701
mmubench
finetune_retain
5,508
0.866667
0.72
0.533333
112.3125
0.050792
0.043219
mmubench
salun
5,508
0.866667
0.78
0.533333
21.484375
0.003207
0.352784

🧠 Multimodal Unlearning Evaluation Benchmark

📌 Overview

This dataset provides evaluation outputs for studying metric inconsistency in multimodal machine unlearning.

It supports reproducibility of results in:

Metric Unreliability in Multimodal Machine Unlearning (NeurIPS 2026)


📊 Contents

File Description
📄 multimodal_results.json Results on VQA benchmarks (MLLMU-Bench, UnLOK-VQA, MMUBench)
📄 unimodal_results.json CIFAR-10 baseline results
⚖️ uqs_weights.json Learned weights for Unified Quality Score (UQS)
🏆 ranking_table.json Method rankings across metrics
📈 analysis_results.json Correlation and disagreement analysis
🔍 kr_pilot_results.json Knowledge Recoverability (KR) pilot results
🤖 blip2_minimal_summary.json Cross-architecture validation (BLIP-2)

🎯 Purpose

This benchmark evaluates five standard unlearning metrics:

  • Forget Accuracy (FA)
  • Retain Accuracy (RA)
  • Membership Inference Attack (MIA)
  • Activation Distance (AD)
  • JS Divergence (JS)

⚠️ Key finding:

These metrics produce conflicting rankings and do not measure knowledge recoverability (KR).


⚙️ Usage

All results in the paper can be reproduced directly from these files.

Example:

import json

with open("multimodal_results.json") as f:
    data = json.load(f)


📚 Source Datasets (Not Included)

This benchmark builds on:

MLLMU-Bench
UnLOK-VQA
MMUBench
CIFAR-10

These datasets are not redistributed here. Please refer to their original sources.

⚖️ License

This dataset is released under the CC-BY-4.0 License.

⚠️ Notes
This dataset contains evaluation outputs, not raw training data
Designed for benchmarking and reproducibility
Prepared to support anonymous peer review
🔗 Citation
Anonymous. Metric Unreliability in Multimodal Machine Unlearning. NeurIPS 2026.

---
Downloads last month
55
Free AI Image Generator No sign-up. Instant results. Open Now