Distinct cortical spatial representations learned along disparate visual pathways.

1. function [S,U,S_history, U_history] = sparse_coding_by_LCA(...
2. X, A, lambda, thresh_type, eta, s_max, n_iter, history_flag, S_past, U_past)
3. % Implement sparse coding using local competition algorithm, i.e. LCA (Rozzel et al. 2008)
4. % X: input matrix where each column is one example of the input
5. % A: the columns of A are basis vectors; the number of columns are the
6. % number of input neurons
7. % lambda: threshold of thresholding function
8. % tau: time constant
9. % n_iter: maximum number of iteration
10. % thresh_type: thresh_type of thresholding function
11. % history_flag: record the trajectories or not
12. % S_past, U_past: previous firing rates and membrane potentials
14. % s_max = 300; % Maximum of neuronal response
16. S_history = []; % Dynamics of the firing rates; each row is the responses of all neurons for the current iteration
17. U_history = []; % Dynamics of the membrane potentials
19. W = A' * A - eye(size(A, 2));
20. U_init = A' * X; % Initial membrane potentials of the neuron
22. if exist('S_past','var') && exist('U_past','var')
23. S = S_past;
24. U = U_past;
25. else
26. U = zeros(size(U_init));
27. S = zeros(size(U_init));
28. end
30. % The process of computing neuronal responses
31. for i = 1 : n_iter
32. % Save the history (trajectory) of model dynamics
33. if exist('history_flag','var') && (history_flag == 1)
34. S_history(i,:) = S(:, 1); % Record the history of the first input
35. U_history(i,:) = U(:, 1); % Record the history of the first input
36. end
38. % Compute the membrane potential, U, of the neuron
39. delta_U = eta * (U_init - U - W*S);
40. U = U + delta_U;
42. % Get firing rate, S, of the neuron by thresholding the membrane, U
43. if isequal(thresh_type, 'soft')
44. S = wthresh(U, 's', lambda);
45. elseif isequal(thresh_type, 'hard')
46. S = wthresh(U, 'h', lambda);
47. elseif isequal(thresh_type,'hard-non-negative')
48. S = max(wthresh(U, 'h', lambda), 0);
49. elseif isequal(thresh_type, 'soft-non-negative')
50. S = max(U - lambda, 0);
51. elseif isequal(thresh_type, 'sigmoid')
52. alpha = 0;% [0,1]
53. gamma = 5;
54. lambda = 1;
55. S = (U - alpha*lambda) ./ (1 + exp(-gamma * (U-lambda)));
56. else
57. error('Not a suitable threshold type. Check thresh_type');
58. end
59. S = min(S, s_max);
60. end

sparse_coding_by_LCA.m at commit 2c30bbf, no license · at the source