Over days and weeks, neurons in mammalian sensorimotor cortex have been found to continually change their activity patterns during performance of a learned sensorimotor task, with no detectable change in behaviour. This challenges classical theories of neural circuit function and memory, which posit that stable engrams underlie stable learned behavior [1,2]. Using existing experimental data we show that a simple linear readout can accurately recover behavioural variables, and that fixed linear weights can approximately decode behaviour over many days, despite significant changes in neural tuning. This implies that an appreciable fraction of ongoing activity reconfiguration occurs in an approximately linear subspace of population activity. We quantify the amount of additional plasticity that would be required to compensate for reconfiguration, and show that a biologically plausible local learning rule can achieve good decoding accuracy with physiologically achievable rates of synaptic plasticity.

Scientists use mathematical modeling as a tool for understanding and predicting the properties of complex physical systems. In highly parametrized models there often exist relationships between parameters over which model predictions are identical, or nearly identical. These are known as structural or practical unidentifiabilities, respectively. They are hard to diagnose and make reliable parameter estimation from data impossible. They furthermore imply the existence of an underlying model simplification. We describe a scalable method for detecting unidentifiabilities, as well as the functional relations defining them, for generic models. This allows for model simplification, and appreciation of which parameters (or functions thereof) cannot be estimated from data. Our algorithm can identify features such as redundant mechanisms and fast time-scale subsystems, as well as the regimes in parameter space over which such approximations are valid. We base our algorithm on a quantification of regional parametric sensitivity that we call 'multiscale sloppiness'. Traditionally, the link between parametric sensitivity and the conditioning of the parameter estimation problem is made locally, through the Fisher information matrix. This is valid in the regime of infinitesimal measurement uncertainty. We demonstrate the duality between multiscale sloppiness and the geometry of confidence regions surrounding parameter estimates made where measurement uncertainty is non-negligible. Further theoretical relationships are provided linking multiscale sloppiness to the likelihood-ratio test. From this, we show that a local sensitivity analysis (as typically done) is insufficient for determining the reliability of parameter estimation, even with simple (non)linear systems. Our algorithm can provide a tractable alternative. We finally apply our methods to a large-scale, benchmark systems biology model of necrosis factor (NF)-κB, uncovering unidentifiabilities.

Over days and weeks, neural activity representing an animal's position and movement in sensorimotor cortex has been found to continually reconfigure or 'drift' during repeated trials of learned tasks, with no obvious change in behavior. This challenges classical theories which assume stable engrams underlie stable behavior. However, it is not known whether this drift occurs systematically, allowing downstream circuits to extract consistent information. Analyzing long-term calcium imaging recordings from posterior parietal cortex in mice (Mus musculus), we show that drift is systematically constrained far above chance, facilitating a linear weighted readout of behavioural variables. However, a significant component of drift continually degrades a fixed readout, implying that drift is not confined to a null coding space. We calculate the amount of plasticity required to compensate drift independently of any learning rule, and find that this is within physiologically achievable bounds. We demonstrate that a simple, biologically plausible local learning rule can achieve these bounds, accurately decoding behavior over many days.

The use of mathematical models in the sciences often involves the estimation of unknown parameter values from data. Sloppiness provides information about the uncertainty of this task. In this paper, we develop a precise mathematical foundation for sloppiness and define rigorously its key concepts, such as 'model manifold', in relation to concepts of structural identifiability. We redefine sloppiness conceptually as a comparison between the premetric on parameter space induced by measurement noise and a reference metric. This opens up the possibility of alternative quantification of sloppiness, beyond the standard use of the Fisher Information Matrix, which assumes that parameter space is equipped with the usual Euclidean metric and the measurement error is infinitesimal. Applications include parametric statistical models, explicit time dependent models, and ordinary differential equation models.

How does the size of a neural circuit influence its learning performance? Larger brains tend to be found in species with higher cognitive function and learning ability. Intuitively, we expect the learning capacity of a neural circuit to grow with the number of neurons and synapses. We show how adding apparently redundant neurons and connections to a network can make a task more learnable. Consequently, large neural circuits can either devote connectivity to generating complex behaviors or exploit this connectivity to achieve faster and more precise learning of simpler behaviors. However, we show that in a biologically relevant setting where synapses introduce an unavoidable amount of noise, there is an optimal size of network for a given task. Above the optimal network size, the addition of neurons and synaptic connections starts to impede learning performance. This suggests that the size of brain circuits may be constrained by the need to learn efficiently with unreliable synapses and provides a hypothesis for why some neurological learning deficits are associated with hyperconnectivity. Our analysis is independent of specific learning rules and uncovers fundamental relationships between learning rate, task performance, network size, and intrinsic noise in neural circuits.