Decoding, as its name suggests is the process of retrieving some information from some 'code'. Code itself is a broad term, but in the field of neuroscience, researchers usually assume that the activity of neurons underly behaviors. Because such activity is complex (millions of neurons firing at a wide range of timescales, often in a stochastic manner), behavior is often believed to be encoded by (or embedded in) neuronal activity. This rationale usually assumes that external stimuli and behavior alone drive neuronal activity. Unfortunately, things are much more complex, and brain activity can be driven by internal factors (brain activity during sleep is a good example).
From the paper 'A probabilistic framework for decoding behavior from calcium imaging data'. You can also refer to the Matlab version of this notebook here: www.github.com/etterguillaume/CaImDecoding
Original article: https://www.frontiersin.org/articles/10.3389/fncir.2020.00019/full
Github source code: www.github.com/etterguillaume/CaImDecoding
Colab notebook:
Figure 5. Rationale for testing significance. After computing the actual neuronal tuning curve, we perform the same analysis on permuted data. A p-value can then be derived, and represents the portion of shuffled surrogate that showed values exceeded our actual data. Finally, it is possible to threshold the original tuning curve so as to display only significant data.
Figure 6. Rationale for establishing confidence intervals.