arXiv subjects

Peter Latham

Publications and source records attributed to Peter Latham.

11 recordsLinked to original sources

arXiv2023

Meta-Learning the Inductive Biases of Simple Neural Circuits

Training data is always finite, making it unclear how to generalise to unseen situations. But, animals do generalise, wielding Occam's razor to select a parsimonious explanation of their observations. How they do this is called their inductive bias, and it is implicitly built into the operation of animals' neural circuits. This relationship between an observed circuit and its inductive bias is a useful explanatory window for neuroscience, allowing design choices to be understood normatively. However, it is generally very difficult to map circuit structure to inductive bias. Here, we present a neural network tool to bridge this gap. The tool meta-learns the inductive bias by learning functions that a neural circuit finds easy to generalise, since easy-to-generalise functions are exactly those the circuit chooses to explain incomplete data. In systems with analytically known inductive bias, i.e. linear and kernel regression, our tool recovers it. Generally, we show it can flexibly extract inductive biases from supervised learners, including spiking neural networks, and show how it could be applied to real animals. Finally, we use our tool to interpret recent connectomic data illustrating its intended use: understanding the role of circuit features through the resulting inductive bias.

q-bio.NC↗

arXiv2022

Restricting admissible representations to fixed-point subgroups

Given a $p$-adic group $G$ equipped with an action of a finite group $Γ\subset\mathrm{Aut}_F(\mathbf{G})$, and a reductive fixed-point subgroup $G^Γ$, we establish a relationship between constructions of types for these two groups due to Yu, Kim--Yu and Fintzen, and generalizes the relationship between general linear and classical groups observed by Stevens. As an application, given a parabolically induced or supercuspidal representation $π$ of $G$, we explicitly identify a number of the inertial equivalence classes occurring in the representation $π|_{G^{[Γ]}}$.

math.RT↗

arXiv2021

Encoding priors in the brain: a reinforcement learning model for mouse decision making

In two-alternative forced choice tasks, prior knowledge can improve performance, especially when operating near the psychophysical threshold. For instance, if subjects know that one choice is much more likely than the other, they can make that choice when evidence is weak. A common hypothesis for these kinds of tasks is that the prior is stored in neural activity. Here we propose a different hypothesis: the prior is stored in synaptic strengths. We study the International Brain Laboratory task, in which a grating appears on either the right or left side of a screen, and a mouse has to move a wheel to bring the grating to the center. The grating is often low in contrast which makes the task relatively difficult, and the prior probability that the grating appears on the right is either 80% or 20%, in (unsignaled) blocks of about 50 trials. We model this as a reinforcement learning task, using a feedforward neural network to map states to actions, and adjust the weights of the network to maximize reward, learning via policy gradient. Our model uses an internal state that stores an estimate of the grating and confidence, and follows Bayesian updates, and can switch between engaged and disengaged states to mimic animal behavior. This model reproduces the main experimental finding - that the psychometric curve with respect to contrast shifts after a block switch in about 10 trials. Also, as seen in the experiments, in our model the difference in neuronal activity in the right and left blocks is small - it is virtually impossible to decode block structure from activity on single trials if noise is about 2%. The hypothesis that priors are stored in weights is difficult to test, but the technology to do so should be available in the not so distant future.

q-bio.NC↗

arXiv2021

Typical representations, parabolic induction and the inertial local Langlands correspondence

We prove a result which provides a link between the decomposition of parabolically induced representations and the Bushnell--Kutzko theory of typical representations. As an application, we show that there exists a well-defined inertial Langlands correspondence which respects the monodromy action of L-parameters, under some standard conjectures regarding the local Langlands correspondence. To allow for potential applications of this inertial Langlands correspondence, we also provide a complete construction of the set of typical representations, giving a parametrization of these in terms of the structure of the Bruhat--Tits building of $G$.

math.RT↗

arXiv2020

Typical representations via fixed point sets in Bruhat--Tits buildings

For an essentially tame supercuspidal representation $π$ of a connected reductive $p$-adic group $G$, we establish two distinct and complementary sufficient conditions for the irreducible components of its restriction to a maximal compact subgroup to occur in a representation of $G$ which is not inertially equivalent to $π$. These two results are further formulated in terms of the geometry of the Bruhat-Tits building of $G$ and its fixed points under the action of certain tori. The consequence is a set of broadly applicable tools for addressing the branching rules of $π$ and the unicity of $[G,π]_G$-types.

math.RT↗

arXiv2018

On the unicity of types for tame toral supercuspidal representations

For tame arbitrary-length toral, also called positive regular, supercuspidal representations of a simply connected and semisimple $p$-adic group $G$, constructed as per Adler-Yu, we determine which components of their restriction to a maximal compact subgroup are types. We give conditions under which there is a unique such component, and then present a class of examples for which there is not, disproving the strong version of the conjecture of unicity of types on maximal compact open subgroups. We restate the unicity conjecture, and prove it holds for the groups and representations under consideration under a mild condition on depth.

math.RT↗

arXiv2017

The unicity of types for depth-zero supercuspidal representations

We establish the unicity of types for depth-zero supercuspidal representations of an arbitrary $p$-adic group $G$, showing that each depth-zero supercuspidal representation of $G$ contains a unique conjugacy class of typical representations of maximal compact subgroups of $G$. As a corollary, we obtain an inertial Langlands correspondence for these representations, via the Langlands correspondence of DeBacker and Reeder.

math.RT↗

arXiv2017

Sparse connectivity for MAP inference in linear models using sister mitral cells

Sensory processing is hard because the variables of interest are encoded in spike trains in a relatively complex way. A major goal in sensory processing is to understand how the brain extracts those variables. Here we revisit a common encoding model in which variables are encoded linearly. Although there are typically more variables than neurons, this problem is still solvable because only a small number of variables appear at any one time (sparse prior). However, previous solutions usually require all-to-all connectivity, inconsistent with the sparse connectivity seen in the brain. Here we propose a principled algorithm that provably reaches the MAP inference solution but using sparse connectivity. Our algorithm is inspired by the mouse olfactory bulb, but our approach is general enough to apply to other modalities; in addition, it should be possible to extend it to nonlinear encoding models.

q-bio.NC↗

arXiv2016

On the unicity of types in special linear groups

Let $F$ be a non-archimedean local field. We show that any representation of a maximal compact subgroup of $\mathbf{SL}_N(F)$ which is typical for an essentially tame supercuspidal representation must be induced from a Bushnell--Kutzko maximal simple type. From this, we explicitly count and describe the conjugacy classes of such typical representations, and give an explicit description of an inertial Langlands correspondence for essentially tame irreducible $N$-dimensional projective representations of the Weil group of $F$.

math.RT↗

arXiv2015

Unicity of types for supercuspidal representations of p-adic SL(2)

We consider the question of unicity of types on maximal compact subgroups for supercuspidal representations of $\mathbf{SL}_2$ over a nonarchimedean local field of odd residual characteristic. We introduce the notion of an archetype as the $\mathbf{SL}_2$-conjugacy class of a typical representation of a maximal compact subgroup, and go on to show that any archetype in $\mathbf{SL}_2$ is restricted from one in $\mathbf{GL}_2$. From this it follows that any archetype must be induced from a Bushnell--Kutzko type. Given a supercuspidal representation $π$, we give an additional explicit description of the number of archetypes admitted by $π$ in terms of its ramification. We also describe a relationship between archetypes for $\mathbf{GL}_2$ and $\mathbf{SL}_2$ in terms of $L$-packets, and deduce an inertial Langlands correspondence for $\mathbf{SL}_2$.

math.RT↗

arXiv2008

Pairwise maximum entropy models for studying large biological systems: when they can and when they can't work

One of the most critical problems we face in the study of biological systems is building accurate statistical descriptions of them. This problem has been particularly challenging because biological systems typically contain large numbers of interacting elements, which precludes the use of standard brute force approaches. Recently, though, several groups have reported that there may be an alternate strategy. The reports show that reliable statistical models can be built without knowledge of all the interactions in a system; instead, pairwise interactions can suffice. These findings, however, are based on the analysis of small subsystems. Here we ask whether the observations will generalize to systems of realistic size, that is, whether pairwise models will provide reliable descriptions of true biological systems. Our results show that, in most cases, they will not. The reason is that there is a crossover in the predictive power of pairwise models: If the size of the subsystem is below the crossover point, then the results have no predictive power for large systems. If the size is above the crossover point, the results do have predictive power. This work thus provides a general framework for determining the extent to which pairwise models can be used to predict the behavior of whole biological systems. Applied to neural data, the size of most systems studied so far is below the crossover point.

q-bio.QM↗