anonymous-researcher-X0006 commited on
Commit
fb74731
·
verified ·
1 Parent(s): 070e1f4

Update README

Browse files
Files changed (1) hide show
  1. README.md +14 -10
README.md CHANGED
@@ -31,12 +31,17 @@ Each composite crest is paired with:
31
  - an English translation,
32
  - a non-linguistic program code over the generator factors.
33
 
34
- Because every crest is synthesized from a known triple of generator factors
35
- container `C`, modifier `R`, motif `M` KamonBench supports direct factor
36
  metrics, controlled factor-pair recombination splits, counterfactual motif-
37
- sensitivity tests under fixed containermodifier contexts, and linear probes
38
- of factor accessibility from frozen representations. See the accompanying
39
- paper for details and baselines.
 
 
 
 
 
40
 
41
  ## Quick start
42
 
@@ -103,8 +108,8 @@ Each Croissant record in the `images` record set has these fields:
103
  For program-label experiments, the same images are paired with non-linguistic
104
  codes for the container (`C:NNN`), modifier (`X:N`), and motif (`M:NNN`); the
105
  three `*_holdout.json` Croissant variants reassign splits so that whole factor
106
- combinations `(C, M)`, `(R, M)`, `(C, R, M)` are absent from training while
107
- the underlying primitive tokens still appear individually in training.
108
 
109
  ## Recombination splits
110
 
@@ -132,9 +137,8 @@ crests.
132
 
133
  The dataset is released under
134
  [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/);
135
- see `LICENSE.txt` for the full legal code. The companion code (under
136
- [`anon-submittere/KamonBench`](https://github.com/anon-submittere/KamonBench)
137
- during anonymous review) is released under the MIT License.
138
 
139
  The component images bundled with KamonBench (one isolated motif per
140
  composite and one container per contained composite) are repackaged in PNG
 
31
  - an English translation,
32
  - a non-linguistic program code over the generator factors.
33
 
34
+ Because every crest is synthesized from a known triple of generator factors
35
+ (container `C`, modifier `R`, motif `M`), KamonBench supports direct factor
36
  metrics, controlled factor-pair recombination splits, counterfactual motif-
37
+ sensitivity tests under fixed (container, modifier) contexts, and linear
38
+ probes of factor accessibility from frozen representations. See the
39
+ accompanying paper for details and baselines.
40
+
41
+ The companion code (package, training and evaluation pipelines, and the
42
+ generator) lives at
43
+ [`anon-submittere/KamonBench`](https://github.com/anon-submittere/KamonBench)
44
+ during anonymous review.
45
 
46
  ## Quick start
47
 
 
108
  For program-label experiments, the same images are paired with non-linguistic
109
  codes for the container (`C:NNN`), modifier (`X:N`), and motif (`M:NNN`); the
110
  three `*_holdout.json` Croissant variants reassign splits so that whole factor
111
+ combinations (`(C, M)`, `(R, M)`, or `(C, R, M)`) are absent from training,
112
+ while the underlying primitive tokens still appear individually in training.
113
 
114
  ## Recombination splits
115
 
 
137
 
138
  The dataset is released under
139
  [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/);
140
+ see `LICENSE.txt` for the full legal code. The companion code is released
141
+ under the MIT License.
 
142
 
143
  The component images bundled with KamonBench (one isolated motif per
144
  composite and one container per contained composite) are repackaged in PNG