-
Inference with correlated priors using sisters cells
Authors:
Sina Tootoonian,
Andreas T. Schaefer
Abstract:
A common view of sensory processing is as probabilistic inference of latent causes from receptor activations. Standard approaches often assume these causes are a priori independent, yet real-world generative factors are typically correlated. Representing such structured priors in neural systems poses architectural challenges, particularly when direct interactions between units representing latent…
▽ More
A common view of sensory processing is as probabilistic inference of latent causes from receptor activations. Standard approaches often assume these causes are a priori independent, yet real-world generative factors are typically correlated. Representing such structured priors in neural systems poses architectural challenges, particularly when direct interactions between units representing latent causes are biologically implausible or computationally expensive. Inspired by the architecture of the olfactory bulb, we propose a novel circuit motif that enables inference with correlated priors without requiring direct interactions among latent cause units. The key insight lies in using sister cells: neurons receiving shared receptor input but connected differently to local interneurons. The required interactions among latent units are implemented indirectly through their connections to the sister cells, such that correlated connectivity implies anti-correlation in the prior and vice versa. We use geometric arguments to construct connectivity that implements a given prior and to bound the number of causes for which such priors can be constructed. Using simulations, we demonstrate the efficacy of such priors for inference in noisy environments and compare the inference dynamics to those experimentally observed. Finally, we show how, under certain assumptions on latent representations, the prior used can be inferred from sister cell activations. While biologically grounded in the olfactory system, our mechanism generalises to other natural and artificial sensory systems and may inform the design of architectures for efficient inference under correlated latent structure.
△ Less
Submitted 20 May, 2025;
originally announced May 2025.
-
Emergent mechanisms for long timescales depend on training curriculum and affect performance in memory tasks
Authors:
Sina Khajehabdollahi,
Roxana Zeraati,
Emmanouil Giannakakis,
Tim Jakob Schäfer,
Georg Martius,
Anna Levina
Abstract:
Recurrent neural networks (RNNs) in the brain and in silico excel at solving tasks with intricate temporal dependencies. Long timescales required for solving such tasks can arise from properties of individual neurons (single-neuron timescale, $τ$, e.g., membrane time constant in biological neurons) or recurrent interactions among them (network-mediated timescale). However, the contribution of each…
▽ More
Recurrent neural networks (RNNs) in the brain and in silico excel at solving tasks with intricate temporal dependencies. Long timescales required for solving such tasks can arise from properties of individual neurons (single-neuron timescale, $τ$, e.g., membrane time constant in biological neurons) or recurrent interactions among them (network-mediated timescale). However, the contribution of each mechanism for optimally solving memory-dependent tasks remains poorly understood. Here, we train RNNs to solve $N$-parity and $N$-delayed match-to-sample tasks with increasing memory requirements controlled by $N$ by simultaneously optimizing recurrent weights and $τ$s. We find that for both tasks RNNs develop longer timescales with increasing $N$, but depending on the learning objective, they use different mechanisms. Two distinct curricula define learning objectives: sequential learning of a single-$N$ (single-head) or simultaneous learning of multiple $N$s (multi-head). Single-head networks increase their $τ$ with $N$ and are able to solve tasks for large $N$, but they suffer from catastrophic forgetting. However, multi-head networks, which are explicitly required to hold multiple concurrent memories, keep $τ$ constant and develop longer timescales through recurrent connectivity. Moreover, we show that the multi-head curriculum increases training speed and network stability to ablations and perturbations, and allows RNNs to generalize better to tasks beyond their training regime. This curriculum also significantly improves training GRUs and LSTMs for large-$N$ tasks. Our results suggest that adapting timescales to task requirements via recurrent interactions allows learning more complex objectives and improves the RNN's performance.
△ Less
Submitted 30 October, 2024; v1 submitted 22 September, 2023;
originally announced September 2023.
-
Analysis of animal-related electric outages using species distribution models and community science data
Authors:
Mei-Ling E. Feng,
Olukunle O. Owolabi,
Toryn L. J. Schafer,
Sanhita Sengupta,
Lan Wang,
David S. Matteson,
Judy P. Che-Castaldo,
Deborah A. Sunter
Abstract:
Animal-related outages (AROs) are a prevalent form of outages in electrical distribution systems. Animal-infrastructure interactions vary across focal species and regions, underlining the need to study the animal-outage relationship in more species and diverse systems. Animal activity has been used as an indicator of reliability in the electrical grid system and to describe temporal patterns in AR…
▽ More
Animal-related outages (AROs) are a prevalent form of outages in electrical distribution systems. Animal-infrastructure interactions vary across focal species and regions, underlining the need to study the animal-outage relationship in more species and diverse systems. Animal activity has been used as an indicator of reliability in the electrical grid system and to describe temporal patterns in AROs. However, these ARO models have been limited by a lack of available estimates of species activity, instead approximating activity based on seasonal and weather patterns in animal-related outage records and characteristics of broad taxonomic groups, e.g., squirrels. We highlight publicly available resources to fill the ecological data gap that is limiting joint analyses between ecology and energy sectors. Species distribution models (SDMs), a common technique to model the distribution of a species across geographic space and time, paired with data sourced from eBird, a community science database for bird observations, provided us with species-specific estimates of activity to model spatio-temporal patterns of AROs. These flexible, species-specific estimates can allow future animal-indicators of grid reliability to be investigated in more diverse regions and ecological communities, providing a better understanding of the variation that exists in animal-outage relationship. AROs were best modeled by accounting for multiple outage-prone species activity patterns and their unique relationships with seasonality and habitat availability. Different species were important for modeling outages in different landscapes and seasons depending on their distribution and migration behavior. We recommend that future models of AROs include species-specific activity data that account for the diverse spectrum of spatio-temporal activity patterns that outage-prone animals exhibit.
△ Less
Submitted 22 December, 2021;
originally announced December 2021.
-
Clustering Future Scenarios Based on Predicted Range Maps
Authors:
Matthew Davidow,
Cory Merow,
Judy Che-Castaldo,
Toryn Schafer,
Marie-Christine Duker,
Derek Corcoran,
David Matteson
Abstract:
Predictions of biodiversity trajectories under climate change are crucial in order to act effectively in maintaining the diversity of species. In many ecological applications, future predictions are made under various global warming scenarios as described by a range of different climate models. The outputs of these various predictions call for a reliable interpretation. We propose a interpretable…
▽ More
Predictions of biodiversity trajectories under climate change are crucial in order to act effectively in maintaining the diversity of species. In many ecological applications, future predictions are made under various global warming scenarios as described by a range of different climate models. The outputs of these various predictions call for a reliable interpretation. We propose a interpretable and flexible two step methodology to measure the similarity between predicted species range maps and cluster the future scenario predictions utilizing a spectral clustering technique. We find that clustering based on ecological impact (predicted species range maps) is mainly driven by the amount of warming. We contrast this with clustering based only on predicted climate features, which is driven mainly by climate models. The differences between these clusterings illustrate that it is crucial to incorporate ecological information to understand the relevant differences between climate models. The findings of this work can be used to better synthesize forecasts of biodiversity loss under the wide spectrum of results that emerge when considering potential future biodiversity loss.
△ Less
Submitted 17 July, 2022; v1 submitted 18 January, 2021;
originally announced January 2021.
-
Poly-Sarcosine and Poly(ethylene-glycol) interactions with proteins investigated using molecular dynamics simulations
Authors:
Giovanni Settanni,
Timo Schäfer,
Christian Muhl,
Matthias Barz,
Friederike Schmid
Abstract:
Nanoparticles coated with hydrophilic polymers often show a reduction in unspecific interactions with the biological environment, which improves their biocompatibility. The molecular determinants of this reduction are not very well understood yet, and their knowledge may help improving nanoparticle design. Here we address, using molecular dynamics simulations, the interactions of human serum album…
▽ More
Nanoparticles coated with hydrophilic polymers often show a reduction in unspecific interactions with the biological environment, which improves their biocompatibility. The molecular determinants of this reduction are not very well understood yet, and their knowledge may help improving nanoparticle design. Here we address, using molecular dynamics simulations, the interactions of human serum albumin, the most abundant serum protein, with two promising hydrophilic polymers used for the coating of therapeutic nanoparticles, poly(ethylene-glycol) and poly-sarcosine. By simulating the protein immersed in a polymer-water mixture, we show that the two polymers have a very similar affinity for the protein surface, both in terms of the amount of polymer adsorbed and also in terms of the type of amino acids mainly involved in the interactions. We further analyze the kinetics of adsorption and how it affects the polymer conformations. Minor differences between the polymers are observed in the thickness of the adsorption layer, that are related to the different degree of flexibility of the two molecules. In comparison poly-alanine, an isomer of poly-sarcosine known to self-aggregate and induce protein aggregation, shows a significantly larger affinity for the protein surface than PEG and PSar, which we show to be related not to a different patterns of interactions with the protein surface, but to the different way the polymer interacts with water.
△ Less
Submitted 29 October, 2018;
originally announced October 2018.
-
Automated Image Analysis of Hodgkin lymphoma
Authors:
Alexander Schmitz,
Tim Schäfer,
Hendrik Schäfer,
Claudia Döring,
Jörg Ackermann,
Norbert Dichter,
Sylvia Hartmann,
Martin-Leo Hansmann,
Ina Koch
Abstract:
Hodgkin lymphoma is an unusual type of lymphoma, arising from malignant B-cells. Morphological and immunohistochemical features of malignant cells and their distribution differ from other cancer types. Based on systematic tissue image analysis, computer-aided exploration can provide new insights into Hodgkin lymphoma pathology. In this paper, we report results from an image analysis of CD30 immuno…
▽ More
Hodgkin lymphoma is an unusual type of lymphoma, arising from malignant B-cells. Morphological and immunohistochemical features of malignant cells and their distribution differ from other cancer types. Based on systematic tissue image analysis, computer-aided exploration can provide new insights into Hodgkin lymphoma pathology. In this paper, we report results from an image analysis of CD30 immunostained Hodgkin lymphoma tissue section images. To the best of our knowledge, this is the first systematic application of image analysis to a set of tissue sections of Hodgkin lymphoma. We have implemented an automatic procedure to handle and explore image data in Aperio's SVS format. We use pre-processing approaches on a down-scaled image to separate the image objects from the background. Then, we apply a supervised classification method to assign pixels to predefined classes. Our pre-processing method is able to separate the tissue content of images from the image background. We analyzed three immunohistologically defined groups, non-lymphoma and the two most common forms of Hodgkin lymphoma, nodular sclerosis and mixed cellularity type. We found that nodular sclerosis and non-lymphoma images exhibit different amounts of CD30 stain, whereas mixed cellularity type exhibits a large variance and overlaps with the other groups. The results can be seen as a first step to computationally identify tumor regions in the images. This allows us to focus on these regions when performing computationally expensive tasks like object detection in the high-resolution image.
△ Less
Submitted 14 September, 2012;
originally announced September 2012.
