Researchers at MIT’s Computer Science and AI Laboratory (CSAIL) have developed a new technique called liquid neural networks (LNNs) to solve the problems of scaling up neural networks and collecting large amounts of labelled training data. LNNs are time-continuous RNNs that process data sequentially, keep the memory of past inputs, and adjust their behaviours based on new inputs. Daniela Rus, the director of CSAIL, recently demonstrated the use of LNNs in running a self-driving autonomous vehicle, showing that with just 19 artificial neurons, the attention map became clearer and more focused on the path that it has to take. LNNs can also be leveraged in various tasks such as embodied robotics, as they are causal and understand cause and effect.
Related Posts
