Daniela Rus
Massachusetts Institute of Technology
- Research field
- Robotics, machine learning, compact and continuous-time neural networks
Daniela Rus directs the Computer Science and Artificial Intelligence Laboratory at MIT and works on robots and the learning systems that control them. In a 2021 paper with Ramin Hasani, Mathias Lechner, Alexander Amini, and Radu Grosu she presented liquid time-constant networks, recurrent networks whose state changes continuously in time according to a simple differential equation in which the rate of change, the time constant, is itself set by the inputs. The paper proves that the states of such a network stay bounded, measures its expressivity, and shows that it beats the LSTM and other recurrent models on several time-series benchmarks. The unit presents it after the recurrent update, as the question what a recurrent network becomes when time is continuous.
In the masterclass
Section 3, after the recurrent update and the vanishing gradient, presents her group's continuous-time recurrent neuron.
The research is presented in Unit 10: Embeddings and Sequence Models, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.