The Korean Brain Neuroinformatics Research Program (1998-2008) has two
goals, i.e., to understand information processing mechanisms in biological
brains and to develop intelligent machines with human-like functions based
on these mechanisms. We are now developing an integrated hardware and
software platform for the brain-like intelligent systems, i.e., Artificial
Brain. With two microphones, two cameras (or retina chips), and one speaker,
the Artificial Brain looks like a human head, and has the functions of
vision, auditory, cognition, and behavior.

The Artificial Brain may be trained to work for specific applications, and
the OfficeMate is our choice of the application test-bed. Similar to office
secretaries the OfficeMate will help users for office jobs such as
scheduling, telephone calls, data search, and document preparation. The
OfficeMate should be able to localize sound in normal office environment,
rotate the head and cameras for visual attention and speech enhancement.
Then it will segment and recognize the face. The lip reading will provide
additional information for robust speech recognition in noisy environment,
and both visual and audio features will be used for the recognition and
representation of “machine emotion.” The OfficeMate will use natural
speech for communications with the human users, while electronic data
communication may be used between OfficeMates.

The Artificial Brain should have sensory modules for human like speech and
visual capabilities, internal state module for the inference, and the output
module for human-like behavioral control. The sensory modules receive audio
and video signals from the environment, and conduct feature extraction and
recognition in the forward path. The backward path conducts top-down
attention, which greatly improves the recognition performance of the
real-world noisy speech and occluded patterns. The fusion of video and audio
signals is also greatly influenced by this backward path. The internal state
module is largely responsible for intelligent functions and has a recurrent
architecture. The recurrent architecture is required to model human-like
emotion and self-esteem. Also, the user adaptation and proactive learning
are performed at this internal state module. The output module generates
human-like behavior with speech synthesizer and facial representation
controller. Also, it provides computer-based services for OfficeMate
applications.

One of the important technologies is to extract features for robust
recognition. We are utilizing unsupervised feature extraction and supervised
feature selection. Several applications are reported including EEG features.