The paper describes a formal model of social network users who have definite sets of interests in different subjects. The users are represented by heterogeneous agents with multiple inputs of different types and multiple outputs of different types (MIMO-agents). Each type corresponds to one of the interests of users. Agents have a cumulative activation function, depending on current external influence from their neighbors and previous network states. If the value of this function at a certain time step is above a specified threshold, the agent becomes active according to one of the types. The choice of this type depends both on his internal structure (personal preferences specified by a vector) and on the proportion of active neighbors of every type. A network of such agents is capable of generating various kinds of complex activity patterns. We consider several examples of activity propagation and show the dependence of stable activity patterns on the parameters of agents. Networks of MIMO-agents with similar properties can be used not only to describe the interaction of users of social networks, but also in modeling the transfer of heterogeneous information in telecommunications networks.