Generalisation experiments.
<p>A. When cells evolve simple forms of plasticity, they are able to generalise, performing better than chance in previously unseen environments (red line represents the information provided (<i>x = y</i>), green line represents the expected performance at random (2<sup><i...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| مؤلفون آخرون: | , |
| منشور في: |
2020
|
| الموضوعات: | |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | <p>A. When cells evolve simple forms of plasticity, they are able to generalise, performing better than chance in previously unseen environments (red line represents the information provided (<i>x = y</i>), green line represents the expected performance at random (2<sup><i>Ne</i></sup><i>-TS</i>)/2; and blue line the degree of matching between the resulting phenotypes and the expected ones for a given function). Notice that blue line runs consistently higher than the green line (random response). B. Similar experiments yield poor performances when cells have to evolve complex forms of plasticity (blue line below the green one, see main text). C. When cells have incomplete information about all possible environments (left), they acquire preferentially simple forms of cell plasticity. Acquiring complex forms of cell plasticity beyond linearly-decomposable functions (see SI) requires full information (blue line). In this panel, the complexity Ω of non-plastic functions was considered to be zero. D. Generalisation experiment under <i>L</i><sub><i>1</i></sub> and <i>L</i><sub><i>2</i></sub> regularization. For this experiment, plastic cells are only exposed to half of the possible environments (<i>TS</i> = 4), that is the point where their inductive bias makes the result most different from randomness. We record how the ability of cells to generalise (to approach the maximum fitness of <i>W</i> = 8) changes as the cost of connections for the GRN increases (see Eq (<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1006811#pcbi.1006811.e008" target="_blank">8</a>)). We do this for <i>L</i><sub><i>1</i></sub> and <i>L</i><sub><i>2</i></sub> regularisation procedures (solid and dashed lines respectively) which favour sparse and weak connections respectively. We also show data for both complex (Ω≈1, red lines) and simple (Ω<0.5, blue lines) plastic responses. The plot shows clearly how including a cost of connections improves drastically the performance of cells in generalisation experiments when cells are asked are evolved towards simple forms of plasticity. For all panels <i>Ne</i> = 3, <i>Ng</i> = 12, <i>n</i> = 30 replicates, standard deviation as boxes, min and max values as error bars.</p> |
|---|