The international neuroscience community is building the first comprehensive atlases of brain cell types to understand how the brain functions from a higher resolution, and more integrated perspective than ever before. In order to build these atlases, subsets of neurons (e.g. serotonergic neurons, prefrontal cortical neurons etc.) are traced in individual brain samples by placing points along dendrites and axons. Then, the traces are mapped to common coordinate systems by transforming the positions of their points, which neglects how the transformation bends the line segments in between. In this work, we apply the theory of jets to describe how to preserve derivatives of neuron traces up to any order. We provide a framework to compute possible error introduced by standard mapping methods, which involves the Jacobian of the mapping transformation. We show how our first order method improves mapping accuracy in both simulated and real neuron traces, though zeroth order mapping is generally adequate in our real data setting. Our method is freely available in our open-source Python package brainlit.
相關內容
A code of length $n$ is said to be (combinatorially) $(\rho,L)$-list decodable if the Hamming ball of radius $\rho n$ around any vector in the ambient space does not contain more than $L$ codewords. We study a recently introduced class of higher order MDS codes, which are closely related (via duality) to codes that achieve a generalized Singleton bound for list decodability. For some $\ell\geq 1$, higher order MDS codes of length $n$, dimension $k$, and order $\ell$ are denoted as $(n,k)$-MDS($\ell$) codes. We present a number of results on the structure of these codes, identifying the `extend-ability' of their parameters in various scenarios. Specifically, for some parameter regimes, we identify conditions under which $(n_1,k_1)$-MDS($\ell_1$) codes can be obtained from $(n_2,k_2)$-MDS($\ell_2$) codes, via various techniques. We believe that these results will aid in efficient constructions of higher order MDS codes. We also obtain a new field size upper bound for the existence of such codes, which arguably improves over the best known existing bound, in some parameter regimes.
In recent years neural networks have achieved impressive results on many technological and scientific tasks. Yet, the mechanism through which these models automatically select features, or patterns in data, for prediction remains unclear. Identifying such a mechanism is key to advancing performance and interpretability of neural networks and promoting reliable adoption of these models in scientific applications. In this paper, we identify and characterize the mechanism through which deep fully connected neural networks learn features. We posit the Deep Neural Feature Ansatz, which states that neural feature learning occurs by implementing the average gradient outer product to up-weight features strongly related to model output. Our ansatz sheds light on various deep learning phenomena including emergence of spurious features and simplicity biases and how pruning networks can increase performance, the "lottery ticket hypothesis." Moreover, the mechanism identified in our work leads to a backpropagation-free method for feature learning with any machine learning model. To demonstrate the effectiveness of this feature learning mechanism, we use it to enable feature learning in classical, non-feature learning models known as kernel machines and show that the resulting models, which we refer to as Recursive Feature Machines, achieve state-of-the-art performance on tabular data.
Feature screening is an important tool in analyzing ultrahigh-dimensional data, particularly in the field of Omics and oncology studies. However, most attention has been focused on identifying features that have a linear or monotonic impact on the response variable. Detecting a sparse set of variables that have a nonlinear or non-monotonic relationship with the response variable is still a challenging task. To fill the gap, this paper proposed a robust model-free screening approach for right-censored survival data by providing a new perspective of quantifying the covariate effect on the restricted mean survival time, rather than the routinely used hazard function. The proposed measure, based on the difference between the restricted mean survival time of covariate-stratified and overall data, is able to identify comprehensive types of associations including linear, nonlinear, non-monotone, and even local dependencies like change points. This approach is highly interpretable and flexible without any distribution assumption. The sure screening property is established and an iterative screening procedure is developed to address multicollinearity between high-dimensional covariates. Simulation studies are carried out to demonstrate the superiority of the proposed method in selecting important features with a complex association with the response variable. The potential of applying the proposed method to handle interval-censored failure time data has also been explored in simulations, and the results have been promising. The method is applied to a breast cancer dataset to identify potential prognostic factors, which reveals potential associations between breast cancer and lymphoma.
Temporal modeling is crucial for multi-frame human pose estimation. Most existing methods directly employ optical flow or deformable convolution to predict full-spectrum motion fields, which might incur numerous irrelevant cues, such as a nearby person or background. Without further efforts to excavate meaningful motion priors, their results are suboptimal, especially in complicated spatiotemporal interactions. On the other hand, the temporal difference has the ability to encode representative motion information which can potentially be valuable for pose estimation but has not been fully exploited. In this paper, we present a novel multi-frame human pose estimation framework, which employs temporal differences across frames to model dynamic contexts and engages mutual information objectively to facilitate useful motion information disentanglement. To be specific, we design a multi-stage Temporal Difference Encoder that performs incremental cascaded learning conditioned on multi-stage feature difference sequences to derive informative motion representation. We further propose a Representation Disentanglement module from the mutual information perspective, which can grasp discriminative task-relevant motion signals by explicitly defining useful and noisy constituents of the raw motion features and minimizing their mutual information. These place us to rank No.1 in the Crowd Pose Estimation in Complex Events Challenge on benchmark dataset HiEve, and achieve state-of-the-art performance on three benchmarks PoseTrack2017, PoseTrack2018, and PoseTrack21.
Active inference is a theory of perception, learning and decision making, which can be applied to neuroscience, robotics, and machine learning. Recently, reasearch has been taking place to scale up this framework using Monte-Carlo tree search and deep learning. The goal of this activity is to solve more complicated tasks using deep active inference. First, we review the existing literature, then, we progresively build a deep active inference agent. For two agents, we have experimented with five definitions of the expected free energy and three different action selection strategies. According to our experiments, the models able to solve the dSprites environment are the ones that maximise rewards. Finally, we compare the similarity of the representation learned by the layers of various agents using centered kernel alignment. Importantly, the agent maximising reward and the agent minimising expected free energy learn very similar representations except for the last layer of the critic network (reflecting the difference in learning objective), and the variance layers of the transition and encoder networks. We found that the reward maximising agent is a lot more certain than the agent minimising expected free energy. This is because the agent minimising expected free energy always picks the action down, and does not gather enough data for the other actions. In contrast, the agent maximising reward, keeps on selecting the actions left and right, enabling it to successfully solve the task. The only difference between those two agents is the epistemic value, which aims to make the outputs of the transition and encoder networks as close as possible. Thus, the agent minimising expected free energy picks a single action (down), and becomes an expert at predicting the future when selecting this action. This makes the KL divergence between the output of the transition and encoder networks small.
To enhance the ability to find credible evidence in news articles, we propose a novel task of expert recommendation, which aims to identify trustworthy experts on a specific news topic. To achieve the aim, we describe the construction of a novel NewsQuote dataset consisting of 24,031 quote-speaker pairs that appeared on a COVID-19 news corpus. We demonstrate an automatic pipeline for speaker and quote extraction via a BERT-based Question Answering model. Then, we formulate expert recommendations as document retrieval task by retrieving relevant quotes first as an intermediate step for expert identification, and expert retrieval by directly retrieving sources based on the probability of a query conditional on a candidate expert. Experimental results on NewsQuote show that document retrieval is more effective in identifying relevant experts for a given news topic compared to expert retrieval
Many major questions in the theory of evolutionary dynamics can in a meaningful sense be mapped to analyses of stochastic trajectories in game theoretic contexts. Often the approach is to analyze small numbers of distinct populations and/or to assume dynamics occur within a regime of population sizes large enough that deterministic trajectories are an excellent approximation of reality. The addition of ecological factors, termed "eco-evolutionary dynamics", further complicates the dynamics and results in many problems which are intractable or impractically messy for current theoretical methods. However, an analogous but underexplored approach is to analyze these systems with an eye primarily towards uncertainty in the models themselves. In the language of researchers in Reinforcement Learning and adjacent fields, a Partially Observable Markov Process. Here we introduce a duality which maps the complexity of accounting for both ecology and individual genotypic/phenotypic types onto a problem of accounting solely for underlying information-theoretic computations rather than drawing physical boundaries which do not change the computations. Armed with this equivalence between computation and the relevant biophysics, which we term Taak-duality, we attack the problem of "directed evolution" in the form of a Partially Observable Markov Decision Process. This provides a tractable case of studying eco-evolutionary trajectories of a highly general type, and of analyzing questions of potential limits on the efficiency of evolution in the directed case.
Motivated by an application from geodesy, we introduce a novel clustering problem which is a $k$-center (or k-diameter) problem with a side constraint. For the side constraint, we are given an undirected connectivity graph $G$ on the input points, and a clustering is now only feasible if every cluster induces a connected subgraph in $G$. We call the resulting problems the connected $k$-center problem and the connected $k$-diameter problem. We prove several results on the complexity and approximability of these problems. Our main result is an $O(\log^2{k})$-approximation algorithm for the connected $k$-center and the connected $k$-diameter problem. For Euclidean metrics and metrics with constant doubling dimension, the approximation factor of this algorithm improves to $O(1)$. We also consider the special cases that the connectivity graph is a line or a tree. For the line we give optimal polynomial-time algorithms and for the case that the connectivity graph is a tree, we either give an optimal polynomial-time algorithm or a $2$-approximation algorithm for all variants of our model. We complement our upper bounds by several lower bounds.
Data in Knowledge Graphs often represents part of the current state of the real world. Thus, to stay up-to-date the graph data needs to be updated frequently. To utilize information from Knowledge Graphs, many state-of-the-art machine learning approaches use embedding techniques. These techniques typically compute an embedding, i.e., vector representations of the nodes as input for the main machine learning algorithm. If a graph update occurs later on -- specifically when nodes are added or removed -- the training has to be done all over again. This is undesirable, because of the time it takes and also because downstream models which were trained with these embeddings have to be retrained if they change significantly. In this paper, we investigate embedding updates that do not require full retraining and evaluate them in combination with various embedding models on real dynamic Knowledge Graphs covering multiple use cases. We study approaches that place newly appearing nodes optimally according to local information, but notice that this does not work well. However, we find that if we continue the training of the old embedding, interleaved with epochs during which we only optimize for the added and removed parts, we obtain good results in terms of typical metrics used in link prediction. This performance is obtained much faster than with a complete retraining and hence makes it possible to maintain embeddings for dynamic Knowledge Graphs.
Graph neural networks (GNNs) are a popular class of machine learning models whose major advantage is their ability to incorporate a sparse and discrete dependency structure between data points. Unfortunately, GNNs can only be used when such a graph-structure is available. In practice, however, real-world graphs are often noisy and incomplete or might not be available at all. With this work, we propose to jointly learn the graph structure and the parameters of graph convolutional networks (GCNs) by approximately solving a bilevel program that learns a discrete probability distribution on the edges of the graph. This allows one to apply GCNs not only in scenarios where the given graph is incomplete or corrupted but also in those where a graph is not available. We conduct a series of experiments that analyze the behavior of the proposed method and demonstrate that it outperforms related methods by a significant margin.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191