Certain aspects of the present disclosure provide methods and apparatus for generating neural adaptive behavior, which may be based on neuromodulator-mediated meta-plasticity and/or gain control. In this manner, flexible associations between sensory cues and motor actions are generated, which enable an agent to efficiently gather rewards in a changing environment. One example method generally includes receiving one or more input stimuli; processing the received input stimuli to generate an output signal, wherein the processing is modulated with a first neuromodulation signal generated by a gain control unit; controlling the gain control unit to switch between at least two different neural activity modes, wherein at least one of a level or timing of the first neuromodulation signal generated by the gain control unit is determined based on the neural activity modes; and sending the output signal to an output unit.