1: \begin{abstract}

2: We study with numerical simulation the 

3: possible limit behaviors

4: of synchronous discrete-time deterministic

5: recurrent neural networks composed of $N$ binary neurons 

6: as a function of a network's level of dilution and asymmetry.

7: The network dilution measures the fraction of neuron couples that are connected,

8: and the network asymmetry measures to what extent

9: the underlying connectivity matrix is asymmetric.

10: For each given neural network,

11: we study the dynamical evolution of all the

12: different initial conditions, thus characterizing the full dynamical 

13: landscape

14: without imposing any learning rule.

15: Because of the deterministic dynamics,

16: each trajectory converges to an attractor,

17: that can be either a fixed point or a limit cycle.

18: These attractors form the set of all the possible limit behaviors of the neural network.

19: For each network we then determine the convergence times, the limit cycles' length,

20: the number of attractors, and the sizes of the attractors' basin.

21: We show that there are two network structures that maximize the

22: number of possible limit behaviors.

23: The first optimal network structure is fully-connected and symmetric.

24: On the contrary, the second optimal network structure is highly sparse and 

25: asymmetric.

26: The latter optimal is similar to what observed in different 

27: biological neuronal circuits.

28: %\textcolor{red}{This implies that biological neural network

29: %which function is to produce as many different behavior

30: %as possible in response

31: %to different stimuli tend to be sparse and asymmetric or fully-connected and symmetric.}

32: These observations lead us to hypothesize that %It is reasonable to assume that,

33: independently from any given learning model,

34: an efficient and effective biologic network that stores

35: a number of limit behaviors close to its maximum capacity

36: tends to develop a connectivity structure

37: similar to one of the optimal networks we found.

38: \end{abstract}

39: