go back

Learning

Learning is one of the key features of intelligence: The capability of using prior experience to adapt intelligent behavior to novel situations. This requires some form of memory that may be divided into short-term and long-term memory. The stored information can be used to adapt and synthesize representations, acquire new skills and change values or preferences. This flexibility widens the scope of intelligent systems going beyond the boundaries of fully pre-programmed solutions.

Incremental learning is characterized by the capability to perform experience-based adaptation from a continuous stream of incoming data. Thus, it facilitates an immediate feedback between a learning system, its user(s) and its environment. HRI-EU develops new approaches for the key challenge of incremental learning systems: finding a good compromise between stability and plasticity of the learned representations. Incremental learning is a prerequisite for personalized assistance systems and HRI-EU’s Collaborative Intelligence vision.

 

Memory Models for Non-Stationary Environments

The capability to deal with change is essential, considering the fact that the world is constantly evolving. Old knowledge may become obsolete or even wrong, contradicting the current beliefs. In these dynamic conditions, algorithms clearly need to capture the current situation and then continuously adapt in order to track changes. In particular, they require a mechanism to decide whether past knowledge is still valid.

We propose the Self-Adjusting Memory (SAM), a new architecture to integrate past and current knowledge in an innovative way. It is based on interlinking the short-term memory (STM) and the long-term memory (LTM). While the STM contains the current concept, the LTM accumulates the information of all past concepts. A consistency between both memories is continuously maintained by selectively filtering contradicting instances. Past knowledge is compressed whenever the upper memory bound is reached. This leads to to an adaptive level of abstraction, making SAM well suited for lifelong learning scenarios like smart home, intelligent life assist or mobility applications.



For more information

V. Losing, B. Hammer, H. Wersing, “Tackling Heterogeneous Concept Drift with the Self Adjusting Memory (SAM)”, Knowledge and Information Systems, 2018

Search