Language models (LMs) have proven to be powerful tools for psycholinguistic research, but most prior work has focused on purely behavioural measures (e.g., surprisal comparisons). At the same time, research in model interpretability has begun to illuminate the abstract causal mechanisms shaping LM behavior. To help bring these strands of research closer together, we introduce CausalGym. We adapt and expand the SyntaxGym suite of tasks to benchmark the ability of interpretability methods to causally affect model behaviour. To illustrate how CausalGym can be used, we study the pythia models (14M--6.9B) and assess the causal efficacy of a wide range of interpretability methods, including linear probing and distributed alignment search (DAS). We find that DAS outperforms the other methods, and so we use it to study the learning trajectory of two difficult linguistic phenomena in pythia-1b: negative polarity item licensing and filler--gap dependencies. Our analysis shows that the mechanism implementing both of these tasks is learned in discrete stages, not gradually.

Training Artificial Neural Networks poses a challenging and critical problem in machine learning. Despite the effectiveness of gradient-based learning methods, such as Stochastic Gradient Descent (SGD), in training neural networks, they do have several limitations. For instance, they require differentiable activation functions, and cannot optimize a model based on several independent non-differentiable loss functions simultaneously; for example, the F1-score, which is used during testing, can be used during training when a gradient-free optimization algorithm is utilized. Furthermore, the training in any DNN can be possible with a small size of the training dataset. To address these concerns, we propose an efficient version of the gradient-free Coordinate Search (CS) algorithm, an instance of General Pattern Search methods, for training neural networks. The proposed algorithm can be used with non-differentiable activation functions and tailored to multi-objective/multi-loss problems. Finding the optimal values for weights of ANNs is a large-scale optimization problem. Therefore instead of finding the optimal value for each variable, which is the common technique in classical CS, we accelerate optimization and convergence by bundling the weights. In fact, this strategy is a form of dimension reduction for optimization problems. Based on the experimental results, the proposed method, in some cases, outperforms the gradient-based approach, particularly, in situations with insufficient labeled training data. The performance plots demonstrate a high convergence rate, highlighting the capability of our suggested method to find a reasonable solution with fewer function calls. As of now, the only practical and efficient way of training ANNs with hundreds of thousands of weights is gradient-based algorithms such as SGD or Adam. In this paper we introduce an alternative method for training ANN.

Uncertainty estimation is a crucial aspect of deploying dependable deep learning models in safety-critical systems. In this study, we introduce a novel and efficient method for deterministic uncertainty estimation called Discriminant Distance-Awareness Representation (DDAR). Our approach involves constructing a DNN model that incorporates a set of prototypes in its latent representations, enabling us to analyze valuable feature information from the input data. By leveraging a distinction maximization layer over optimal trainable prototypes, DDAR can learn a discriminant distance-awareness representation. We demonstrate that DDAR overcomes feature collapse by relaxing the Lipschitz constraint that hinders the practicality of deterministic uncertainty methods (DUMs) architectures. Our experiments show that DDAR is a flexible and architecture-agnostic method that can be easily integrated as a pluggable layer with distance-sensitive metrics, outperforming state-of-the-art uncertainty estimation methods on multiple benchmark problems.

Previous studies in predicting crash risk primarily associated the number or likelihood of crashes on a road segment with traffic parameters or geometric characteristics of the segment, usually neglecting the impact of vehicles' continuous movement and interactions with nearby vehicles. Advancements in communication technologies have empowered driving information collected from surrounding vehicles, enabling the study of group-based crash risks. Based on high-resolution vehicle trajectory data, this research focused on vehicle groups as the subject of analysis and explored risk formation and propagation mechanisms considering features of vehicle groups and road segments. Several key factors contributing to crash risks were identified, including past high-risk vehicle-group states, complex vehicle behaviors, high percentage of large vehicles, frequent lane changes within a vehicle group, and specific road geometries. A multinomial logistic regression model was developed to analyze the spatial risk propagation patterns, which were classified based on the trend of high-risk occurrences within vehicle groups. The results indicated that extended periods of high-risk states, increase in vehicle-group size, and frequent lane changes are associated with adverse risk propagation patterns. Conversely, smoother traffic flow and high initial crash risk values are linked to risk dissipation. Furthermore, the study conducted sensitivity analysis on different types of classifiers, prediction time intervalsss and adaptive TTC thresholds. The highest AUC value for vehicle-group risk prediction surpassed 0.93. The findings provide valuable insights to researchers and practitioners in understanding and prediction of vehicle-group safety, ultimately improving active traffic safety management and operations of Connected and Autonomous Vehicles.

Minimax optimization problems have attracted a lot of attention over the past few years, with applications ranging from economics to machine learning. While advanced optimization methods exist for such problems, characterizing their dynamics in stochastic scenarios remains notably challenging. In this paper, we pioneer the use of stochastic differential equations (SDEs) to analyze and compare Minimax optimizers. Our SDE models for Stochastic Gradient Descent-Ascent, Stochastic Extragradient, and Stochastic Hamiltonian Gradient Descent are provable approximations of their algorithmic counterparts, clearly showcasing the interplay between hyperparameters, implicit regularization, and implicit curvature-induced noise. This perspective also allows for a unified and simplified analysis strategy based on the principles of It\^o calculus. Finally, our approach facilitates the derivation of convergence conditions and closed-form solutions for the dynamics in simplified settings, unveiling further insights into the behavior of different optimizers.

In statistical learning theory, a generalization bound usually involves a complexity measure imposed by the considered theoretical framework. This limits the scope of such bounds, as other forms of capacity measures or regularizations are used in algorithms. In this paper, we leverage the framework of disintegrated PAC-Bayes bounds to derive a general generalization bound instantiable with arbitrary complexity measures. One trick to prove such a result involves considering a commonly used family of distributions: the Gibbs distributions. Our bound stands in probability jointly over the hypothesis and the learning sample, which allows the complexity to be adapted to the generalization gap as it can be customized to fit both the hypothesis class and the task.

In the domain of causal inference research, the prevalent potential outcomes framework, notably the Rubin Causal Model (RCM), often overlooks individual interference and assumes independent treatment effects. This assumption, however, is frequently misaligned with the intricate realities of real-world scenarios, where interference is not merely a possibility but a common occurrence. Our research endeavors to address this discrepancy by focusing on the estimation of direct and spillover treatment effects under two assumptions: (1) network-based interference, where treatments on neighbors within connected networks affect one's outcomes, and (2) non-random treatment assignments influenced by confounders. To improve the efficiency of estimating potentially complex effects functions, we introduce an novel active learning approach: Active Learning in Causal Inference with Interference (ACI). This approach uses Gaussian process to flexibly model the direct and spillover treatment effects as a function of a continuous measure of neighbors' treatment assignment. The ACI framework sequentially identifies the experimental settings that demand further data. It further optimizes the treatment assignments under the network interference structure using genetic algorithms to achieve efficient learning outcome. By applying our method to simulation data and a Tencent game dataset, we demonstrate its feasibility in achieving accurate effects estimations with reduced data requirements. This ACI approach marks a significant advancement in the realm of data efficiency for causal inference, offering a robust and efficient alternative to traditional methodologies, particularly in scenarios characterized by complex interference patterns.

A new model for generating survival trajectories and data based on applying an autoencoder of a specific structure is proposed. It solves three tasks. First, it provides predictions in the form of the expected event time and the survival function for a new generated feature vector on the basis of the Beran estimator. Second, the model generates additional data based on a given training set that would supplement the original dataset. Third, the most important, it generates a prototype time-dependent trajectory for an object, which characterizes how features of the object could be changed to achieve a different time to an event. The trajectory can be viewed as a type of the counterfactual explanation. The proposed model is robust during training and inference due to a specific weighting scheme incorporating into the variational autoencoder. The model also determines the censored indicators of new generated data by solving a classification task. The paper demonstrates the efficiency and properties of the proposed model using numerical experiments on synthetic and real datasets. The code of the algorithm implementing the proposed model is publicly available.

The performance of spectral clustering relies on the fluctuations of the entries of the eigenvectors of a similarity matrix, which has been left uncharacterized until now. In this letter, it is shown that the signal $+$ noise structure of a general spike random matrix model is transferred to the eigenvectors of the corresponding Gram kernel matrix and the fluctuations of their entries are Gaussian in the large-dimensional regime. This CLT-like result was the last missing piece to precisely predict the classification performance of spectral clustering. The proposed proof is very general and relies solely on the rotational invariance of the noise. Numerical experiments on synthetic and real data illustrate the universality of this phenomenon.

Trees are convenient models for obtaining explainable predictions on relatively small datasets. Although there are many proposals for the end-to-end construction of such trees in supervised learning, learning a tree end-to-end for clustering without labels remains an open challenge. As most works focus on interpreting with trees the result of another clustering algorithm, we present here a novel end-to-end trained unsupervised binary tree for clustering: Kauri. This method performs a greedy maximisation of the kernel KMeans objective without requiring the definition of centroids. We compare this model on multiple datasets with recent unsupervised trees and show that Kauri performs identically when using a linear kernel. For other kernels, Kauri often outperforms the concatenation of kernel KMeans and a CART decision tree.