An exciting day of talks and discussions: December 18, 2026

Titles  & Abstracts

​(in order of presentation)
 

Peter Latham                 Time and task management in complex environments

​

 In daily life, we are confronted with a variety of tasks: eating, sleeping, working, socializing, etc. A key feature of these tasks is that the amount of satisfaction we get -- the reward rate -- varies with time. In particular, it typically decays over time (the first bite of a meal brings us much more satisfaction than the 50th), while at the same time the reward rates in tasks we're not engaged in rise (the longer we go without sleeping the more rewarding sleep becomes). The question we address here is: how do we schedule our tasks to maximize long term reward? This is an instance of optimal foraging, but the depletion and  recovery of reward rates makes it significantly more complicated. And somewhat non-intuitive: we show that agents should spend a significant amount of time in unrewarded tasks (which could explain why we sleep); if there's no cost of switching they should switch infinitely often; and if there is a cost to switching, the time spend in tasks typically scales as the dead time to the 1/3 power (an easily testable experimental
prediction). We end by speculating on the implications of these results for time management in the real world.

​

​

​Magnus Richardson        Spatiotemporal coarse-graining of activity within neurons and across networks

​

Compared to point-neuron models, there are relatively few analytical results on the role of space in the integration of synaptic input. Here, a cascade of partial differential equations is derived for the voltage statistics of a minimal dendritic model driven by temporally filtered synapses. Despite its simplicity, the model exhibits rich dynamic properties including a fast response to afferent drive and morphological resonances. At the level of population activity, a space-time rate-based model is obtained from coarse-graining a network of spiking neurons. For biophysically reasonable choices of neuron model and spatial coupling, a known delay partial differential equation with parameters linked to the underlying neuron model can be found, providing a tractable starting point for other coarse-graining approaches to network activity.

​

​

​​​

Rafal Bogacz                  Predictive dendrites as a foundation for biological learning
 

A fundamental challenge for neuroscience is to identify laws of synaptic plasticity that both agree with experimental data and enable networks of neurons to learn complex tasks. We develop and evaluate a promising candidate law stating that neurons adjust synaptic strengths to minimize the mismatch between somatic and dendritic activity. We demonstrate that models based on this rule effectively adjust synaptic weights across layers to learn simple non-linear tasks and can be extended to improve learning on more complex tasks. Moreover, for spiking neurons, we show that minimization of mismatch between spiking activity and dendritic inputs leads to integrate-and-fire dynamics and spike time dependent plasticity. Furthermore, the proposed model reproduces the dependence of plasticity on initial weights and on the stimulation patterns including spike pairs and triples. Consequently, the model captures fundamental experimentally observed properties of neurons and provides a foundation for studying how their networks can learn complex tasks.

​​
 

Arnd Roth                       Brain-wide dendritic signals exhibit circuit-specific behavioural modulation
​

 Dendrites are ubiquitous features of neuronal morphology. Dendritic signals – in particular backpropagating action potentials (bAPs), which convey a retrograde signal about axonal output to dendritic synapses – have been proposed to play key roles in synaptic integration and plasticity. However, how these signals are engaged during behaviour across neuronal types and brain regions remains largely unexplored. We develop a robust framework to identify extracellular signatures of bAPs in large-scale Neuropixels recordings, backed by simulations, and apply it to brain-wide data from 139 mice performing an adaptive visual decision-making task. We detect signatures of bAPs across 72 brain regions, spanning cortical and subcortical structures including frontal, visual, motor, and somatosensory cortex, as well as striatum, midbrain and thalamus. Single-spike analyses show that dendritic backpropagation is a dynamic signal shaped by recent spike timing and spike-history structure. Across the brain, bAPs are modulated by behavioural state. During perceptual decision making, bAP dynamics carry structured representations of sensory, choice, and feedback variables, with temporal organization that differs from corresponding somatic firing-rate representations. Finally, latent behavioural states reorganize dendritic signaling across distributed circuits, revealing a continuous representational landscape linking intrinsic physiology, internal state and task variables. By revealing a dimension of neuronal signalling beyond somatic spiking, our findings suggest that dendritic excitability represents an important component of behaviourally relevant computations in biological neural circuits.
                                        

​

​

​Mark Humphries           Cortical control of rhythmic and discrete arm movements

Arm movements are rhythmic, discrete, or some combination of the two. Conflicting evidence supports each of two possible solutions for how motor cortex controls them: that either it uses the same strategy for controlling rhythmic and discrete movements or different strategies for distinct movement types. Using recurrent neural network modelling and multi-unit recordings during an arm-cycling task, we
show that primate motor cortex uses both solutions. Primary motor cortex (M1) dynamics converge to the same limit-cycle when executing both movement types. In contrast, supplementary motor area (SMA) dynamics diverge according to the type of the upcoming movement before reaching a helical spiral. Our results propose SMA is critical for generating discrete movements by controlling when and how M1
dynamics reach their limit cycle.  (With Andrea Colins Rodriguez & Romulo Fuentes.)
 

​
​
​​

Vladimir Litvak               Low Effective Dimensionality of the Human Functional Connectome Does Not           

                                            Invalidate Lesion Network Mapping

​​

Lesion network mapping (LNM) has gained prominence as a method for relating sets of non‑overlapping brain lesions to the symptoms they produce. LNM maps symptom‑specific functional circuits and suggests potential targets for invasive and non‑invasive neuromodulation. Recently, Van den Heuvel et al. argued that the low effective dimensionality of the human connectome causes lesion network maps to converge on a common pattern, the connectome degree map, and concluded that the method cannot resolve distinct symptom-specific circuits. We show that, although lesion network maps are low-dimensional and may be biased towards the dominant mode of the connectome, they still recover their ground truth networks well. Moreover, we propose a modification of the connectome that reduces this bias by half without reducing overall recovery. Our results, therefore, reaffirm the validity of lesion network mapping, while acknowledging the important limitations identified by its critics. (With Thomas Zaugg.)



 

Massimiliano di Luca   Estimating when: Temporal inference in perception and coordinated action.
                                            Low Effective Dimensionality of the Human Functional Connectome Does Not           

                                            Invalidate Lesion Network Mapping

As sensory streams of information unfold, humans continuously update estimates of when events have occurred. I will present psychophysical and computational work showing how uncertain sensory evidence and expectations jointly shape perceived event time. In regular sequences, a recursive Bayesian model captures timing biases by integrating each posterior distribution into the next temporal prior. I will contrast these expectation effects with multisensory recalibration following repeated audiovisual asynchrony. Finally, I will turn to musical ensembles, where a Kalman-filter model has been used to estimate time-varying phase-correction gains and is employed to create a music rehearsal tool. Together, these studies reveal complementary forms of temporal adjustment across perception, virtual interaction and interpersonal coordination.