mwang@watmath.UUCP (mwang) (05/15/84)
_D_E_P_A_R_T_M_E_N_T _O_F _C_O_M_P_U_T_E_R _S_C_I_E_N_C_E
_U_N_I_V_E_R_S_I_T_Y _O_F _W_A_T_E_R_L_O_O
_C_O_M_P_U_T_E_R _S_C_I_E_N_C_E _C_O_L_L_O_Q_U_I_U_M
- Wednesday, May 23, 1984.
Dr. L. Rendell of the University of Guelph will speak
on ``Efficient Generalization Learning in Search.''
TIME: 3:30 PM
ROOM: MC 5158
ABSTRACT
Generalization learning is inductive inference, ac-
quisition of conceptual knowledge. It can be accom-
plished efficiently by structuring statistics obtained
from observation of a given primary task. Such an ap-
proach has wide application in noisy environments.
The probabilistic learning system (PLS) is currently
being studied using the domain of search in state-space
problems and games. Feature space clustering,
knowledge accumulation, and regression have resulted in
economical discovery of locally optimal evaluation
functions. Adding an upper layer of (parallel) learn-
ing increases stability and power; adding lower layers
may allow feature formation from elementary data (i.e.
full inductive capability).
Some unifying concepts and methods are suggested for
efficient generalization learning.
Coffee and refreshments will be served at 3:00 PM.
May 15, 1984