From Nature ( 4 JANUARY 2018 | VOL 553 | )
Deep learning sharpens views of cells and genes
Neural networks are making biological images easier to process.
Amy Maxmen
............
Other machine-learning connoisseurs in biology have set their sights on new frontiers, now that convolutional neural networks are taking flight for image processing. “Imaging is important, but so is chemistry and molecular data,” says Alex Wolf, a computational biologist at the German Research Center for Environmental Health in Neuherberg. Wolf hopes to tweak neural networks so that they can analyse gene expression. “I think there will be a very big breakthrough in the next few years,” he says, “that allows biologists to apply neural networks much more broadly.”
artificial neural networks are good for imaging and for the prediction of the data. But initially you have to train the network.
Neural network is a black box approach to modeling, It is applied to modeling a process or system when fundamental knowledge about the system is very limited, Thus, neural network can be used to predict certain outputs given certain input values. In this manner it does not help explain the detailed undertsanding of the process, i.e., the fundamental why the outputs are as such.
so, we can use one black box (i.e. neural network) to try to predict behavior of another black box (for example cellular response to certain inputs)
Another very interesting problem is adversarial input. (e.g. https://arxiv.org/pdf/1412.6572.pdf )
But it is more of a general problem than a specific one for NNs. You can sum it up as: We can use the gradient to maipulate the input in a way that we are able to fool the network, by adding a very small portion (less than 1% can be sufficient) of of the specific "noise" (it's not really noise but to humans it looks like it) to the input. This "noise" also generalizes over various models, even discret ones, according to Goodfellow. In general we know too few about those high dimensional spaces and Deep-Learning in theory, to know precisely why this is happening, but Goodfellow suggests that this is due to the high linearity in the models.
- Badges
- Science topic
- Similar topics
- Biology