Neural network
A layered model that learns patterns from examples. Like successive filters, the layers transform the input into useful representations. Performance depends on the task, the data and the evaluation — there is no automatic superiority.
FROID Psique · Glossary
Ten concepts for understanding FROID Psique. The analogies make reading easier; they do not replace definitions or evidence of validity.
A layered model that learns patterns from examples. Like successive filters, the layers transform the input into useful representations. Performance depends on the task, the data and the evaluation — there is no automatic superiority.
Retrieval-augmented generation. The search brings relevant content to support the model's answer: like consulting a library before writing. The answer can still be wrong.
Multimodal Power Index. Despite its historical name, in the current composition it organizes the magnitude of acoustic deviations. It is a reading of variation, not a mental health score.
Multimodal Deviation Index. It preserves the average direction of vocal deviations, distinguishing increases and decreases in energy. It does not measure conflict between speech and face.
Bands that organize vocal energy. Think of the channels of an equalizer, not twelve emotions. There is no validated association between each band and a feeling.
Groupings of compatible movement configurations. They work like categories in a catalog; they do not determine the person's emotional state.
FACS describes movements in Action Units. The product uses approximate estimates: 16 fields, 15 with mappings conditioned on the measurements received, and AU23 with no ascertainment. It is not certified coding.
Mel-frequency cepstral coefficients: a mathematical representation of spectral characteristics of speech. In the product, MFCC7 and MFCC9 are tracked measures, not isolated diagnoses.
The product uses this name for exploratory indicators associated with vocal analysis. The described 5–12 Hz range must not be presented as a validated clinical biomarker or as a diagnosis.
A property of a set in which each record shares combinations of selected attributes with at least k−1 others. It does not mean zero risk of re-identification; its application depends on the data and on the conditions of the analysis. In Data-Froid, an aggregate query is only answered for groups with k ≥ 7 records; below that it is refused. The floor limits the query; it does not make re-identification impossible.