Combining the Perceptron Algorithm with Logarithmic Simulated Annealing

摘要: We present results of computational experiments with an extension the Perceptron algorithm by a special type simulated annealing. The annealing procedure employs logarithmic cooling schedule c(k)eΓ/ln(k+2), where Γ is parameter that depends on underlying configuration space. For sample sets S n-dimensional vectors generated randomly chosen polynomials w1·x1a1+···+wn·xnangt, we try to approximate positive and negative examples linear threshold functions. approximations are computed both classical our schedules. ne256,…, 1024 aie3,…, 7, outperforms about 15% when size sufficiently large. was according estimations maximum escape depth from local minima associated energy landscape.