http://dl.gi.de/handle/20.500.12116/11112 2026-07-22T22:11:25Z 2026-07-22T22:11:25Z Walter, Oliver Haeb-Umbach, Reinhold Mokbel, Bassam Paassen, Benjamin Hammer, Barbara http://dl.gi.de/handle/20.500.12116/11485 2018-03-20T10:33:09Z 2015-01-01T00:00:00Z

Autonomous Learning of Representations Walter, Oliver; Haeb-Umbach, Reinhold; Mokbel, Bassam; Paassen, Benjamin; Hammer, Barbara Besides the core learning algorithm itself, one major question in machine learning is how to best encode given training data such that the learning technology can efficiently learn based thereon and generalize to novel data. While classical approaches often rely on a hand coded data representation, the topic of autonomous representation or feature learning plays a major role in modern learning architectures. The goal of this contribution is to give an overview about different principles of autonomous feature learning, and to exemplify two principles based on two recent examples: autonomous metric learning for sequences, and autonomous learning of a deep representation for spoken language, respectively.

2015-01-01T00:00:00Z Hutter, Frank Lücke, Jörg Schmidt-Thieme, Lars http://dl.gi.de/handle/20.500.12116/11487 2018-03-20T10:33:09Z 2015-01-01T00:00:00Z

Beyond Manual Tuning of Hyperparameters Hutter, Frank; Lücke, Jörg; Schmidt-Thieme, Lars The success of hand-crafted machine learning systems in many applications raises the question of making machine learning algorithms more autonomous, i.e., to reduce the requirement of expert input to a minimum. We discuss two strategies towards this goal: (1) automated optimization of hyperparameters (including mechanisms for feature selection, preprocessing, model selection, etc) and (2) the development of algorithms with reduced sets of hyperparameters. Since many research directions (e.g., deep learning), show a tendency towards increasingly complex algorithms with more and more hyperparamters, the demand for both of these strategies continuously increases. We review recent hyperparameter optimization methods and discuss data-driven approaches to avoid the introduction of hyperparameters using unsupervised learning. We end in discussing how these complementary strategies can work hand-in-hand, representing a very promising approach towards autonomous machine learning.

2015-01-01T00:00:00Z Missura, Marcell Behnke, Sven http://dl.gi.de/handle/20.500.12116/11491 2018-03-20T10:33:09Z 2015-01-01T00:00:00Z

Online Learning of Bipedal Walking Stabilization Missura, Marcell; Behnke, Sven Bipedal walking is a complex whole-body motion with inherently unstable dynamics that makes the design of a robust controller particularly challenging. While a walk controller could potentially be learned with the hardware in the loop, the destructive nature of exploratory motions and the impracticality of a high number of required repetitions render most of the existing machine learning methods unsuitable for an online learning setting with real hardware. In a project in the DFG Priority Programme Autonomous Learning, we are investigating ways of bootstrapping the learning process with basic walking skills and enabling a humanoid robot to autonomously learn how to control its balance during walking.

2015-01-01T00:00:00Z Ay, Nihat http://dl.gi.de/handle/20.500.12116/11482 2018-03-20T10:33:09Z 2015-01-01T00:00:00Z

Geometric Design Principles for Brains of Embodied Agents Ay, Nihat I propose a formal model of the sensorimotor loop and discuss corresponding extrinsic embodiment constraints and the intrinsic degrees of freedom. These degrees constitute the basis for adaptation in terms of learning and should therefore be coupled with the embodiment constraints. Notions of sufficiency and embodied universal approximation allow us to formulate principles for such a coupling. This provides a geometric approach to the design of control architectures for embodied agents.

2015-01-01T00:00:00Z