EN 
Українська
Polski
Español
Deutsch
日本語
Italiano
Português (BR)
Français
العربية
Nederlands
中文
한국어
हिन्दी
Türkçe
Tiếng Việt
Русский
Bahasa Indonesia
فارسی
বাংলা
ไทย
Filipino
Kiswahili
اردو
Bahasa Melayu
தமிழ்
עברית
Ελληνικά
ਪੰਜਾਬੀ
አማርኛ
မြန်မာဘာသာ
ខ្មែរ
ქართული
नेपाली
Монгол хэл
Հայերեն
සිංහල
ລາວ
Čeština
Română
Magyar
Български
Svenska
Dansk
Suomi
Norsk
Slovenčina
Hrvatski
Српски
Slovenščina
Lietuvių
Latviešu
Eesti
Íslenska
Shqip
Македонски
Bosanski
Malti
Azərbaycanca
Қазақша
O'zbekcha
Тоҷикӣ
КыргызчаCounterfactual Explanations
Understanding through alternative scenarios is key to effective AI explanation.
Counterfactual explanations (counterfacts) are a powerful method of AI explainability, showing what changes need to be made in the input data to achieve a different outcome from the model. Instead of explaining *why* a model made a decision, counterfacts demonstrate an alternate scenario: "If your income were $5000 higher, the loan would have been approved."
z' = optimize(z for f(decoder(z')) = y_target)
They generate realistic counterfactuals.
These operate within the latent space, allowing for targeted modifications of the input data.
2. Medicine and Healthcare
Critically important applications include:
Diagnosis: "What changes would lead to a different diagnosis?"
Frequently asked questions
What data types are used with DiCE/Alibi?
Data types used include tabular data – DiCE/Alibi, and image data – generative models.
What type of model is typically employed for counterfactual generation?
Typically, differentiable models are used with gradient-based methods, alongside tree-based search approaches.
How does the speed of a counterfactual explanation impact its complexity?
Real-time counterfactuals often rely on simpler methods, while offline explanations require more complex calculations.
How important is the realism of the generated counterfactual?
The realism of a counterfactual is critical when using generative models; however, optimization techniques can sometimes prioritize efficiency over perfect realism.
▶ Try it live
Everything above runs in your browser — open Hash Function Avalanche Visualizer and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.