Providing the neurobiological basis of information processing in higher animals, spiking neural networks must be able to learn a variety of complicated computations, including the generation of appropriate, possibly delayed reactions to inputs and the self-sustained generation of complex activity patterns, e.g. for locomotion. Many such computations require previous building of intrinsic world models. Here we show how spiking neural networks may solve these different tasks. Firstly, we derive constraints under which classes of spiking neural networks lend themselves to substrates of powerful general purpose computing. The networks contain dendritic or synaptic nonlinearities and have a constrained connectivity. We then combine such networks with learning rules for outputs or recurrent connections. We show that this allows to learn even difficult benchmark tasks such as the self-sustained generation of desired low-dimensional chaotic dynamics or memory-dependent computations. Furthermore, we show how spiking networks can build models of external world systems and use the acquired knowledge to control them.

Language processing involves the ability to store and integrate pieces of information in working memory over short periods of time. According to the dominant view, information is maintained through sustained, elevated neural activity. Other work has argued that short-term synaptic facilitation can serve as a substrate of memory. Here we propose an account where memory is supported by intrinsic plasticity that downregulates neuronal firing rates. Single neuron responses are dependent on experience, and we show through simulations that these adaptive changes in excitability provide memory on timescales ranging from milliseconds to seconds. On this account, spiking activity writes information into coupled dynamic variables that control adaptation and move at slower timescales than the membrane potential. From these variables, information is continuously read back into the active membrane state for processing. This neuronal memory mechanism does not rely on persistent activity, excitatory feedback, or synaptic plasticity for storage. Instead, information is maintained in adaptive conductances that reduce firing rates and can be accessed directly without cued retrieval. Memory span is systematically related to both the time constant of adaptation and baseline levels of neuronal excitability. Interference effects within memory arise when adaptation is long lasting. We demonstrate that this mechanism is sensitive to context and serial order which makes it suitable for temporal integration in sequence processing within the language domain. We also show that it enables the binding of linguistic features over time within dynamic memory registers. This work provides a step toward a computational neurobiology of language.

Biologically inspired spiking networks are an important tool to study the nature of computation and cognition in neural systems. In this work, we investigate the representational capacity of spiking networks engaged in an identity mapping task. We compare two schemes for encoding symbolic input, one in which input is injected as a direct current and one where input is delivered as a spatio-temporal spike pattern. We test the ability of networks to discriminate their input as a function of the number of distinct input symbols. We also compare performance using either membrane potentials or filtered spike trains as state variable. Furthermore, we investigate how the circuit behavior depends on the balance between excitation and inhibition, and the degree of synchrony and regularity in its internal dynamics. Finally, we compare different linear methods of decoding population activity onto desired target labels. Overall, our results suggest that even this simple mapping task is strongly influenced by design choices on input encoding, state-variables, circuit characteristics and decoding methods, and these factors can interact in complex ways. This work highlights the importance of constraining computational network models of behavior by available neurobiological evidence.

In language processing, an interpretation is computed incrementally within memory while utterances unfold in time. Here, we investigate the nature of this processing memory in a spiking network model of sentence comprehension. We show that the history dependence of neuronal responses endows circuits of biological neurons with adequate memory to assign semantic roles and resolve binding relations between words in a stream of language input. A neurobiological read-write memory is proposed where short-lived spiking activity encodes information into coupled dynamic variables that move at slower timescales. This state-dependent network does not rely on persistent activity, excitatory feedback, or synaptic plasticity for storage. Instead, information is maintained in adaptive neuronal conductances and can be accessed directly during comprehension without cued retrieval of previous input words. This work provides a step towards a computational neurobiology of language. / (Marder et al.,; Zhang and Linden, ) which has been argued to play an important role in memory formation and maintenance (Titley et al., ). Since language is a neurobiological system, we address this question through the simulation of neurobiologically motivated networks engaged in sentence comprehension. This causal modeling approach, which takes the biophysical substrate of cognitive behavior as a starting point (Koch, ), di ers fundamentally from existing arti cial neural network models of language processing (Elman,; McRae et al., ; Ueno et al., ; Hinaut and Dominey, ; Rabovsky et al.,). These accounts are not modeling spike generation, neuronal adaptation, or the kinetics of synaptic transmission. Hence, they do not simulate the primitives of neural computation (e.g., action potentials and their timing) and are unable to express processing memory as history dependent changes in neuronal excitability. Crucially, these comprehension models have mainly focused on the reproduction of human behavioral data, and data tting is often achieved through powerful, supervised learning methods whose neurobiological basis is unclear (Marblestone et al., ). In contrast, the causal modeling approach pursued here attempts to ground language processing in basic neurobiological design principles. In other words, rather than reproduce a speci c aspect of behavior, our aim is to understand the causal role of these very principles. Compliance with behavioral or neuroimaging data is an independent, future outcome of this approach, not a primary modeling goal. As opposed to previous models of sentence comprehension, parameters in causal models have units of measurement that need to fall within physiological bounds. This places strong constraints on the model space and curbs arbitrary choice available in cognitive modeling. Furthermore, network time corresponds to real, physical time and this allows us to map the temporal dimension of language processing in a principled manner. Modeling the system instead of data will eventually explain how language processing and memory...