Research connected to &quot;machine learning&quot;

AdaS: Adaptive Scheduling of Stochastic Gradients

The choice of step-size used in Stochastic Gradient Descent (SGD) optimization is empirically selected in most training procedures. Moreover, the use of scheduled learning techniques such as Step-Decaying, Cyclical-Learning, and Warmup to tune the step-size requires extensive practical experience--offering limited insight into how the parameters update--and is not consistent across applications. This work attempts to answer a question of interest to both researchers and practitioners, namely \textit{"how much knowledge is gained in iterative training of deep neural networks?"} Answering this question introduces two useful metrics derived from the singular values of the low-rank factorization of convolution layers in deep neural networks. We introduce the notions of \textit{"knowledge gain"} and \textit{"mapping condition"} and propose a new algorithm called Adaptive Scheduling (AdaS) that utilizes these derived metrics to adapt the SGD learning rate proportionally to the rate of change in knowledge gain over successive iterations. Experimentation reveals that, using the derived metrics, AdaS exhibits: (a) faster convergence and superior generalization over e

Matrix Sensing with Kernel Optimal Loss: Robustness and Optimization Landscape

In this paper we study how the choice of loss functions of non-convex optimization problems affects their robustness and optimization landscape, through the study of noisy matrix sensing. In traditional regression tasks, mean squared error (MSE) loss is a common choice, but it can be unreliable for non-Gaussian or heavy-tailed noise. To address this issue, we adopt a robust loss based on nonparametric regression, which uses a kernel-based estimate of the residual density and maximizes the estimated log-likelihood. This robust formulation coincides with the MSE loss under Gaussian errors but remains stable under more general settings. We further examine how this robust loss reshapes the optimization landscape by analyzing the upper-bound of restricted isometry property (RIP) constants for spurious local minima to disappear. Through theoretical and empirical analysis, we show that this new loss excels at handling large noise and remains robust across diverse noise distributions. This work offers initial insights into enhancing the robustness of machine learning tasks through simply changing the loss, guided by an intuitive and broadly applicable analytical framework.

Detecting Suspicious Events in Fast Information Flows

We describe a computational feather-light and intuitive, yet provably efficient algorithm, named HALFADO. HALFADO is designed for detecting suspicious events in a high-frequency stream of complex entries, based on a relatively small number of examples of human judgement. Operating a sufficiently accurate detection system is vital for {\em assisting} teams of human experts in many different areas of the modern digital society. These systems have intrinsically a far-reaching normative effect, and public knowledge of the workings of such technology should be a human right. On a conceptual level, the present approach extends one of the most classical learning algorithms for classification, inheriting its theoretical properties. It however works in a semi-supervised way integrating human and computational intelligence. On a practical level, this algorithm transcends existing approaches (expert systems) by managing and boosting their performance into a single global detector. We illustrate HALFADO's efficacy on two challenging applications: (1) for detecting {\em hate speech} messages in a flow of text messages gathered from a social media platform, and (2) for a Transaction Monitoring System (TMS) in FinTech detecting fraudulent transactions in a stream of financial transactions. This algorithm illustrates that - contrary to popular belief - advanced methods of machine learning need not require neither advanced levels of computation power nor expensive annotation efforts.

Supervised Topological Maps

Controlling the internal representation space of a neural network is a desirable feature because it allows to generate new data in a supervised manner. In this paper we will show how this can be achieved while building a low-dimensional mapping of the input stream, by deriving a generalized algorithm starting from Self Organizing Maps (SOMs). SOMs are a kind of neural network which can be trained with unsupervised learning to produce a low-dimensional discretized mapping of the input space. They can be used for the generation of new data through backward propagation of interpolations made from the mapping grid. Unfortunately the final topology of the mapping space of a SOM is not known before learning, so interpolating new data in a supervised way is not an easy task. Here we will show a variation from the SOM algorithm consisting in constraining the update of prototypes so that it is also a function of the distance of its prototypes from extrinsically given targets in the mapping space. We will demonstrate how such variants, that we will call Supervised Topological Maps (STMs), allow for a supervised mapping where the position of internal representations in the mapping space is de

A Regret Perspective on Online Multiple Testing

Online Multiple Testing (OMT), a fundamental pillar of sequential statistical inference, traditionally evaluates the False Discovery Rate (FDR) and statistical power in isolation, obscuring the highly asymmetric costs of false positives and false negatives in modern automated pipelines. To unify this evaluation, we introduce $\textit{Weighted Regret}$. Under this metric, we prove the $\textit{Duality of Regret Conservation}$: purely deterministic procedures ensuring strict FDR control inevitably incur an $Ω(T)$ linear regret penalty, as threshold depletion during signal-sparse cold starts forces massive false negatives. Tailored for exogenous testing streams, we propose Decoupled-OMT (DOMT) as a baseline-agnostic meta-wrapper. By incorporating a history-decoupled, strictly non-negative random perturbation, DOMT rescues purely deterministic baselines from severe threshold depletion. Crucially, it preserves exact asymptotic safety in stationary environments and rigorously bounds finite-sample error inflation during cold-starts. Guaranteeing zero additional false negatives, it yields an order-optimal $Ω(\sqrt{T})$ regret reduction in bursty environments, with a derived ``Cold-Start Tax'' characterizing the exact phase transition of algorithmic superiority. Experiments validate that DOMT consistently curtails empirical weighted regret, achieving an order-optimal sublinear mitigation of threshold depletion to navigate the non-stationary Pareto frontier.

A Closer Look at Accuracy vs. Robustness

Current methods for training robust networks lead to a drop in test accuracy, which has led prior works to posit that a robustness-accuracy tradeoff may be inevitable in deep learning. We take a closer look at this phenomenon and first show that real image datasets are actually separated. With this property in mind, we then prove that robustness and accuracy should both be achievable for benchmark datasets through locally Lipschitz functions, and hence, there should be no inherent tradeoff between robustness and accuracy. Through extensive experiments with robustness methods, we argue that the gap between theory and practice arises from two limitations of current methods: either they fail to impose local Lipschitzness or they are insufficiently generalized. We explore combining dropout with robust training methods and obtain better generalization. We conclude that achieving robustness and accuracy in practice may require using methods that impose local Lipschitzness and augmenting them with deep learning generalization techniques. Code available at https://github.com/yangarbiter/robust-local-lipschitz

Universal Data Anomaly Detection via Inverse Generative Adversary Network

The problem of detecting data anomaly is considered. Under the null hypothesis that models anomaly-free data, measurements are assumed to be from an unknown distribution with some authenticated historical samples. Under the composite alternative hypothesis, measurements are from an unknown distribution positive distance away from the distribution under the null hypothesis. No training data are available for the distribution of anomaly data. A semi-supervised deep learning technique based on an inverse generative adversary network is proposed.

High-throughput relation extraction algorithm development associating knowledge articles and electronic health records

Objective: Medical relations are the core components of medical knowledge graphs that are needed for healthcare artificial intelligence. However, the requirement of expert annotation by conventional algorithm development processes creates a major bottleneck for mining new relations. In this paper, we present Hi-RES, a framework for high-throughput relation extraction algorithm development. We also show that combining knowledge articles with electronic health records (EHRs) significantly increases the classification accuracy. Methods: We use relation triplets obtained from structured databases and semistructured webpages to label sentences from target corpora as positive training samples. Two methods are also provided for creating improved negative samples by combining positive samples with naïve negative samples. We propose a common model that summarizes sentence information using large-scale pretrained language models and multi-instance attention, which then joins with the concept embeddings trained from the EHRs for relation prediction. Results: We apply the Hi-RES framework to develop classification algorithms for disorder-disorder relations and disorder-location relations. Mill

preprint2015arXiv

Using Machine Learning to Predict the Outcome of English County twenty over Cricket Matches

Cricket betting is a multi-billion dollar market. Therefore, there is a strong incentive for models that can predict the outcomes of games and beat the odds provided by bookers. The aim of this study was to investigate to what degree it is possible to predict the outcome of cricket matches. The target competition was the English twenty over county cricket cup. The original features alongside engineered features gave rise to more than 500 team and player statistics. The models were optimized firstly with team features only and then both team and player features. The performance of the models was tested over individual seasons from 2009 to 2014 having been trained over previous season data in each case. The optimal model was a simple prediction method combined with complex hierarchical features and was shown to significantly outperform a gambling industry benchmark.

Winning solutions and post-challenge analyses of the ChaLearn AutoDL challenge 2019

This paper reports the results and post-challenge analyses of ChaLearn's AutoDL challenge series, which helped sorting out a profusion of AutoML solutions for Deep Learning (DL) that had been introduced in a variety of settings, but lacked fair comparisons. All input data modalities (time series, images, videos, text, tabular) were formatted as tensors and all tasks were multi-label classification problems. Code submissions were executed on hidden tasks, with limited time and computational resources, pushing solutions that get results quickly. In this setting, DL methods dominated, though popular Neural Architecture Search (NAS) was impractical. Solutions relied on fine-tuned pre-trained networks, with architectures matching data modality. Post-challenge tests did not reveal improvements beyond the imposed time limit. While no component is particularly original or novel, a high level modular organization emerged featuring a "meta-learner", "data ingestor", "model selector", "model/learner", and "evaluator". This modularity enabled ablation studies, which revealed the importance of (off-platform) meta-learning, ensembling, and efficient data management. Experiments on heterogeneous module combinations further confirm the (local) optimality of the winning solutions. Our challenge legacy includes an ever-lasting benchmark (http://autodl.chalearn.org), the open-sourced code of the winners, and a free "AutoDL self-service".

preprint2012arXiv

Transfer Learning Using Logistic Regression in Credit Scoring

The credit scoring risk management is a fast growing field due to consumer's credit requests. Credit requests, of new and existing customers, are often evaluated by classical discrimination rules based on customers information. However, these kinds of strategies have serious limits and don't take into account the characteristics difference between current customers and the future ones. The aim of this paper is to measure credit worthiness for non customers borrowers and to model potential risk given a heterogeneous population formed by borrowers customers of the bank and others who are not. We hold on previous works done in generalized gaussian discrimination and transpose them into the logistic model to bring out efficient discrimination rules for non customers' subpopulation. Therefore we obtain several simple models of connection between parameters of both logistic models associated respectively to the two subpopulations. The German credit data set is selected to experiment and to compare these models. Experimental results show that the use of links between the two subpopulations improve the classification accuracy for the new loan applicants.

preprint2013arXiv

Fast Dual Variational Inference for Non-Conjugate LGMs

Latent Gaussian models (LGMs) are widely used in statistics and machine learning. Bayesian inference in non-conjugate LGMs is difficult due to intractable integrals involving the Gaussian prior and non-conjugate likelihoods. Algorithms based on variational Gaussian (VG) approximations are widely employed since they strike a favorable balance between accuracy, generality, speed, and ease of use. However, the structure of the optimization problems associated with these approximations remains poorly understood, and standard solvers take too long to converge. We derive a novel dual variational inference approach that exploits the convexity property of the VG approximations. We obtain an algorithm that solves a convex optimization problem, reduces the number of variational parameters, and converges much faster than previous methods. Using real-world data, we demonstrate these advantages on a variety of LGMs, including Gaussian process classification, and latent Gaussian Markov random fields.

Training Non-Differentiable Networks via Optimal Transport

Neural networks increasingly embed non-differentiable components (spiking neurons, quantized layers, discrete routing, blackbox simulators, etc.) where backpropagation is inapplicable and surrogate gradients introduce bias. We present PolyStep, a gradient-free optimizer that updates parameters using only forward passes. Each step evaluates the loss at structured polytope vertices in a compressed subspace, computes softmax-weighted assignments over the resulting cost matrix, and displaces particles toward low-cost vertices via barycentric projection. This update corresponds to the one-sided limit of a regularized optimal-transport problem, inheriting its geometric structure without Sinkhorn iterations. PolyStep trains genuinely non-differentiable models where existing gradient-free methods collapse to near-random accuracy. On hard-LIF spiking networks we reach 93.4% test accuracy, outperforming all gradient-free baselines by over 60~pp and closing to within 4.4~pp of a surrogate-gradient Adam ceiling. Across four additional non-differentiable architectures (int8 quantization, argmax attention, staircase activations, hard MoE routing) we lead every gradient-free competitor. On MAX-SAT scaling from 100 to 1M variables, we sustain above 92% clause satisfaction while evolution strategies drop 8--12~pp. On RL policy search, we match OpenAI-ES on classical control and retain performance under integer and binary quantization that collapses gradient-based methods. We prove convergence to conservative-stationary points at rate $O(\log T/\sqrt{T})$ on piecewise-smooth losses, upgraded to Clarke-stationary on the headline architectures and extended to the piecewise-constant regime via a hitting-time bound. These rates match the known zeroth-order query-complexity lower bounds that all forward-only methods inherit. Code is available at https://github.com/anindex/polystep.

preprint2014arXiv

Implicit Temporal Differences

In reinforcement learning, the TD($λ$) algorithm is a fundamental policy evaluation method with an efficient online implementation that is suitable for large-scale problems. One practical drawback of TD($λ$) is its sensitivity to the choice of the step-size. It is an empirically well-known fact that a large step-size leads to fast convergence, at the cost of higher variance and risk of instability. In this work, we introduce the implicit TD($λ$) algorithm which has the same function and computational cost as TD($λ$), but is significantly more stable. We provide a theoretical explanation of this stability and an empirical evaluation of implicit TD($λ$) on typical benchmark tasks. Our results show that implicit TD($λ$) outperforms standard TD($λ$) and a state-of-the-art method that automatically tunes the step-size, and thus shows promise for wide applicability.

Explainability-Guided Defense: Attribution-Aware Model Refinement Against Adversarial Data Attacks

The growing reliance on deep learning models in safety-critical domains such as healthcare and autonomous navigation underscores the need for defenses that are both robust to adversarial perturbations and transparent in their decision-making. In this paper, we identify a connection between interpretability and robustness that can be directly leveraged during training. Specifically, we observe that spurious, unstable, or semantically irrelevant features identified through Local Interpretable Model-Agnostic Explanations (LIME) contribute disproportionately to adversarial vulnerability. Building on this insight, we introduce an attribution-guided refinement framework that transforms LIME from a passive diagnostic into an active training signal. Our method systematically suppresses spurious features using feature masking, sensitivity-aware regularization, and adversarial augmentation in a closed-loop refinement pipeline. This approach does not require additional datasets or model architectures and integrates seamlessly into standard adversarial training. Theoretically, we derive an attribution-aware lower bound on adversarial distortion that formalizes the link between explanation alig

preprint2024arXiv

A unified uncertainty-aware exploration: Combining epistemic and aleatory uncertainty

Exploration is a significant challenge in practical reinforcement learning (RL), and uncertainty-aware exploration that incorporates the quantification of epistemic and aleatory uncertainty has been recognized as an effective exploration strategy. However, capturing the combined effect of aleatory and epistemic uncertainty for decision-making is difficult. Existing works estimate aleatory and epistemic uncertainty separately and consider the composite uncertainty as an additive combination of the two. Nevertheless, the additive formulation leads to excessive risk-taking behavior, causing instability. In this paper, we propose an algorithm that clarifies the theoretical connection between aleatory and epistemic uncertainty, unifies aleatory and epistemic uncertainty estimation, and quantifies the combined effect of both uncertainties for a risk-sensitive exploration. Our method builds on a novel extension of distributional RL that estimates a parameterized return distribution whose parameters are random variables encoding epistemic uncertainty. Experimental results on tasks with exploration and risk challenges show that our method outperforms alternative approaches.

Technical Note on Transcription Factor Motif Discovery from Importance Scores (TF-MoDISco) version 0.5.6.5

TF-MoDISco (Transcription Factor Motif Discovery from Importance Scores) is an algorithm for identifying motifs from basepair-level importance scores computed on genomic sequence data. This technical note focuses on version v0.5.6.5. The implementation is available at https://github.com/kundajelab/tfmodisco/tree/v0.5.6.5

Self-supervised Learning for ECG-based Emotion Recognition

We present an electrocardiogram (ECG) -based emotion recognition system using self-supervised learning. Our proposed architecture consists of two main networks, a signal transformation recognition network and an emotion recognition network. First, unlabelled data are used to successfully train the former network to detect specific pre-determined signal transformations in the self-supervised learning step. Next, the weights of the convolutional layers of this network are transferred to the emotion recognition network, and two dense layers are trained in order to classify arousal and valence scores. We show that our self-supervised approach helps the model learn the ECG feature manifold required for emotion recognition, performing equal or better than the fully-supervised version of the model. Our proposed method outperforms the state-of-the-art in ECG-based emotion recognition with two publicly available datasets, SWELL and AMIGOS. Further analysis highlights the advantage of our self-supervised approach in requiring significantly less data to achieve acceptable results.

preprint2015arXiv

A Mathematical Theory for Clustering in Metric Spaces

Clustering is one of the most fundamental problems in data analysis and it has been studied extensively in the literature. Though many clustering algorithms have been proposed, clustering theories that justify the use of these clustering algorithms are still unsatisfactory. In particular, one of the fundamental challenges is to address the following question: What is a cluster in a set of data points? In this paper, we make an attempt to address such a question by considering a set of data points associated with a distance measure (metric). We first propose a new cohesion measure in terms of the distance measure. Using the cohesion measure, we define a cluster as a set of points that are cohesive to themselves. For such a definition, we show there are various equivalent statements that have intuitive explanations. We then consider the second question: How do we find clusters and good partitions of clusters under such a definition? For such a question, we propose a hierarchical agglomerative algorithm and a partitional algorithm. Unlike standard hierarchical agglomerative algorithms, our hierarchical agglomerative algorithm has a specific stopping criterion and it stops with a parti

Variational Inference for Lévy Process-Driven SDEs via Neural Tilting

Modelling extreme events and heavy-tailed phenomena is central to building reliable predictive systems in domains such as finance, climate science, and safety-critical AI. While Lévy processes provide a natural mathematical framework for capturing jumps and heavy tails, Bayesian inference for Lévy-driven stochastic differential equations (SDEs) remains intractable with existing methods: Monte Carlo approaches are rigorous but lack scalability, whereas neural variational inference methods are efficient but rely on Gaussian assumptions that fail to capture discontinuities. We address this tension by introducing a neural exponential tilting framework for variational inference in Lévy-driven SDEs. Our approach constructs a flexible variational family by exponentially reweighting the Lévy measure using neural networks. This parametrization preserves the jump structure of the underlying process while remaining computationally tractable. To enable efficient inference, we develop a quadratic neural parametrization that yields closed-form normalization of the tilted measure, a conditional Gaussian representation for stable processes that facilitates simulation, and symmetry-aware Monte Carlo estimators for scalable optimization. Empirically, we demonstrate that the method accurately captures jump dynamics and yields reliable posterior inference in regimes where Gaussian-based variational approaches fail, on both synthetic and real-world datasets.

Soft-MSM: Differentiable Context-Aware Elastic Alignment for Time Series

Elastic distances like dynamic time warping (DTW) are central to time series machine learning because they compare sequences under local temporal misalignment. Soft-DTW is an adaptation of DTW that can be used as a gradient-based loss by replacing the hard minimum in its dynamic-programming recursion with a smooth relaxation. However, this approach does not directly extend to elastic distances whose transition costs depend on the local alignment context. Move-Split-Merge (MSM) is one such distance: it uses context-aware split and merge penalties and has often outperformed DTW in supervised and unsupervised time series machine learning tasks such as classification and clustering. We introduce Soft-MSM, a smooth relaxation of MSM and an elastic alignment loss with context-aware transition costs. Central to the formulation is a smooth gated surrogate for MSM's piecewise split/merge cost, which enables gradients through both the dynamic-programming recursion and the local transition structure. We derive the forward recursion, backward recursion, soft alignment matrix, closed-form gradient, limiting behaviour, and divergence-corrected formulation. Experiments on 112 UCR datasets show that Soft-MSM gives lower MSM barycentre loss than existing MSM barycentre methods, and yields significantly better clustering and nearest-centroid classification performance than Soft-DTW-based alternatives. An implementation is available in the open-source \texttt{aeon} toolkit.

A Deep-Learning Based Optimization Approach to Address Stop-Skipping Strategy in Urban Rail Transit Lines

Different passenger demand rates in transit stations underscore the importance of adopting operational strategies to provide a demand-responsive service. Aiming at improving passengers' travel time, the present study introduces an advanced data-driven optimization approach to determine the optimal stop-skip pattern in urban rail transit lines. In detail, first, using the time-series smart card data for an entire month, we employ a Long Short-Term Memory (LSTM) deep learning model to predict the station-level demand rates for the peak hour. This prediction is based on four preceding hours and is especially important knowing that the true demand rates of the peak hour are posterior information that can be obtained only after the peak hour operation is finished. Moreover, utilizing a real-time prediction instead of assuming fixed demand rates, allows us to account for unexpected real-time changes which can be detrimental to the subsequent analyses. Then, we integrate the output of the LSTM model as an input to an optimization model with the objective of minimizing patrons' total travel time. Considering the exponential nature of the problem, we propose an Ant Colony Optimization technique to solve the problem in a desirable amount of time. Finally, the performance of the proposed models and the solution algorithm is assessed using real case data. The results suggest that the proposed approach can enhance the performance of the service by improving both passengers' in-vehicle time as well as passengers' waiting time.

BATS: Best Action Trajectory Stitching

The problem of offline reinforcement learning focuses on learning a good policy from a log of environment interactions. Past efforts for developing algorithms in this area have revolved around introducing constraints to online reinforcement learning algorithms to ensure the actions of the learned policy are constrained to the logged data. In this work, we explore an alternative approach by planning on the fixed dataset directly. Specifically, we introduce an algorithm which forms a tabular Markov Decision Process (MDP) over the logged data by adding new transitions to the dataset. We do this by using learned dynamics models to plan short trajectories between states. Since exact value iteration can be performed on this constructed MDP, it becomes easy to identify which trajectories are advantageous to add to the MDP. Crucially, since most transitions in this MDP come from the logged data, trajectories from the MDP can be rolled out for long periods with confidence. We prove that this property allows one to make upper and lower bounds on the value function up to appropriate distance metrics. Finally, we demonstrate empirically how algorithms that uniformly constrain the learned policy to the entire dataset can result in unwanted behavior, and we show an example in which simply behavior cloning the optimal policy of the MDP created by our algorithm avoids this problem.

$β$-Intact-VAE: Identifying and Estimating Causal Effects under Limited Overlap

As an important problem in causal inference, we discuss the identification and estimation of treatment effects (TEs) under limited overlap; that is, when subjects with certain features belong to a single treatment group. We use a latent variable to model a prognostic score which is widely used in biostatistics and sufficient for TEs; i.e., we build a generative prognostic model. We prove that the latent variable recovers a prognostic score, and the model identifies individualized treatment effects. The model is then learned as β-Intact-VAE--a new type of variational autoencoder (VAE). We derive the TE error bounds that enable representations balanced for treatment groups conditioned on individualized features. The proposed method is compared with recent methods using (semi-)synthetic datasets.

The Trap of Trajectory: Towards Understanding and Mitigating Spurious Correlations in Agentic Memory

Agentic memory enables LLMs to persist information beyond a single context window and reuse it in later decisions, but it also introduces a new vulnerability: spurious correlations, where retrieved memory carries miscorrelated evidence and propagates erroneous reasoning into downstream decisions. Despite the widespread use of agentic memory, this risk remains largely underexplored. We address it from two aspects. First, we benchmark several canonical types of spurious patterns identified through causal structure and record them across trajectory-level memory. Diagnosing agentic memory systems on this benchmark reveals that memory improves reasoning on clean inputs but amplifies reliance on spurious patterns when they are present. Second, we propose CAMEL, a plug-and-play calibration method that operates across diverse memory architectures at both write and retrieval time. CAMEL consistently reduces reliance on spurious patterns across all three types while preserving or improving performance on clean inputs and staying robust under adaptive attacks targeting the calibration. Overall, CAMEL offers a principled and lightweight solution toward more reliable agentic memory deployment.

Enhance transferability of adversarial examples with model architecture

Transferability of adversarial examples is of critical importance to launch black-box adversarial attacks, where attackers are only allowed to access the output of the target model. However, under such a challenging but practical setting, the crafted adversarial examples are always prone to overfitting to the proxy model employed, presenting poor transferability. In this paper, we suggest alleviating the overfitting issue from a novel perspective, i.e., designing a fitted model architecture. Specifically, delving the bottom of the cause of poor transferability, we arguably decompose and reconstruct the existing model architecture into an effective model architecture, namely multi-track model architecture (MMA). The adversarial examples crafted on the MMA can maximumly relieve the effect of model-specified features to it and toward the vulnerable directions adopted by diverse architectures. Extensive experimental evaluation demonstrates that the transferability of adversarial examples based on the MMA significantly surpass other state-of-the-art model architectures by up to 40% with comparable overhead.

Rissanen Data Analysis: Examining Dataset Characteristics via Description Length

We introduce a method to determine if a certain capability helps to achieve an accurate model of given data. We view labels as being generated from the inputs by a program composed of subroutines with different capabilities, and we posit that a subroutine is useful if and only if the minimal program that invokes it is shorter than the one that does not. Since minimum program length is uncomputable, we instead estimate the labels' minimum description length (MDL) as a proxy, giving us a theoretically-grounded method for analyzing dataset characteristics. We call the method Rissanen Data Analysis (RDA) after the father of MDL, and we showcase its applicability on a wide variety of settings in NLP, ranging from evaluating the utility of generating subquestions before answering a question, to analyzing the value of rationales and explanations, to investigating the importance of different parts of speech, and uncovering dataset gender bias.

Policy Optimization with Sparse Global Contrastive Explanations

We develop a Reinforcement Learning (RL) framework for improving an existing behavior policy via sparse, user-interpretable changes. Our goal is to make minimal changes while gaining as much benefit as possible. We define a minimal change as having a sparse, global contrastive explanation between the original and proposed policy. We improve the current policy with the constraint of keeping that global contrastive explanation short. We demonstrate our framework with a discrete MDP and a continuous 2D navigation domain.

preprint2013arXiv

Consistent Bounded-Asynchronous Parameter Servers for Distributed ML

In distributed ML applications, shared parameters are usually replicated among computing nodes to minimize network overhead. Therefore, proper consistency model must be carefully chosen to ensure algorithm's correctness and provide high throughput. Existing consistency models used in general-purpose databases and modern distributed ML systems are either too loose to guarantee correctness of the ML algorithms or too strict and thus fail to fully exploit the computing power of the underlying distributed system. Many ML algorithms fall into the category of \emph{iterative convergent algorithms} which start from a randomly chosen initial point and converge to optima by repeating iteratively a set of procedures. We've found that many such algorithms are to a bounded amount of inconsistency and still converge correctly. This property allows distributed ML to relax strict consistency models to improve system performance while theoretically guarantees algorithmic correctness. In this paper, we present several relaxed consistency models for asynchronous parallel computation and theoretically prove their algorithmic correctness. The proposed consistency models are implemented in a di