Despite their ability to represent highly expressive functions, deep learning models seem to find simple solutions that generalize surprisingly well. Spectral bias -- the tendency of neural networks to prioritize learning low frequency functions -- is one possible explanation for this phenomenon, but so far spectral bias has primarily been observed in theoretical models and simplified experiments. In this work, we propose methodologies for measuring spectral bias in modern image classification networks on CIFAR-10 and ImageNet. We find that these networks indeed exhibit spectral bias, and that interventions that improve test accuracy on CIFAR-10 tend to produce learned functions that have higher frequencies overall but lower frequencies in the vicinity of examples from each class. This trend holds across variation in training time, model architecture, number of training examples, data augmentation, and self-distillation. We also explore the connections between function frequency and image frequency and find that spectral bias is sensitive to the low frequencies prevalent in natural images. On ImageNet, we find that learned function frequency also varies with internal class diversity, with higher frequencies on more diverse classes. Our work enables measuring and ultimately influencing the spectral behavior of neural networks used for image classification, and is a step towards understanding why deep models generalize well.

Explainability methods for NLP systems encounter a version of the fundamental problem of causal inference: for a given ground-truth input text, we never truly observe the counterfactual texts necessary for isolating the causal effects of model representations on outputs. In response, many explainability methods make no use of counterfactual texts, assuming they will be unavailable. In this paper, we show that robust causal explainability methods can be created using approximate counterfactuals, which can be written by humans to approximate a specific counterfactual or simply sampled using metadata-guided heuristics. The core of our proposal is the Causal Proxy Model (CPM). A CPM explains a black-box model $\mathcal{N}$ because it is trained to have the same actual input/output behavior as $\mathcal{N}$ while creating neural representations that can be intervened upon to simulate the counterfactual input/output behavior of $\mathcal{N}$. Furthermore, we show that the best CPM for $\mathcal{N}$ performs comparably to $\mathcal{N}$ in making factual predictions, which means that the CPM can simply replace $\mathcal{N}$, leading to more explainable deployed models. Our code is available at https://github.com/frankaging/Causal-Proxy-Model.

Contrastive learning has recently achieved remarkable success in many domains including graphs. However contrastive loss, especially for graphs, requires a large number of negative samples which is unscalable and computationally prohibitive with a quadratic time complexity. Sub-sampling is not optimal and incorrect negative sampling leads to sampling bias. In this work, we propose a meta-node based approximation technique that can (a) proxy all negative combinations (b) in quadratic cluster size time complexity, (c) at graph level, not node level, and (d) exploit graph sparsity. By replacing node-pairs with additive cluster-pairs, we compute the negatives in cluster-time at graph level. The resulting Proxy approximated meta-node Contrastive (PamC) loss, based on simple optimized GPU operations, captures the full set of negatives, yet is efficient with a linear time complexity. By avoiding sampling, we effectively eliminate sample bias. We meet the criterion for larger number of samples, thus achieving block-contrastiveness, which is proven to outperform pair-wise losses. We use learnt soft cluster assignments for the meta-node constriction, and avoid possible heterophily and noise added during edge creation. Theoretically, we show that real world graphs easily satisfy conditions necessary for our approximation. Empirically, we show promising accuracy gains over state-of-the-art graph clustering on 6 benchmarks. Importantly, we gain substantially in efficiency; up to 3x in training time, 1.8x in inference time and over 5x in GPU memory reduction.

This paper describes (R)ules (o)f (T)he (R)oad (A)dvisor, an agent that provides recommended and possible actions to be generated from a set of human-level rules. We describe the architecture and design of RoTRA, both formally and with an example. Specifically, we use RoTRA to formalise and implement the UK "Rules of the Road", and describe how this can be incorporated into autonomous cars such that they can reason internally about obeying the rules of the road. In addition, the possible actions generated are annotated to indicate whether the rules state that the action must be taken or that they only recommend that the action should be taken, as per the UK Highway Code (Rules of The Road). The benefits of utilising this system include being able to adapt to different regulations in different jurisdictions; allowing clear traceability from rules to behaviour, and providing an external automated accountability mechanism that can check whether the rules were obeyed in some given situation. A simulation of an autonomous car shows, via a concrete example, how trust can be built by putting the autonomous vehicle through a number of scenarios which test the car's ability to obey the rules of the road. Autonomous cars that incorporate this system are able to ensure that they are obeying the rules of the road and external (legal or regulatory) bodies can verify that this is the case, without the vehicle or its manufacturer having to expose their source code or make their working transparent, thus allowing greater trust between car companies, jurisdictions, and the general public.

Various Non-negative Matrix factorization (NMF) based methods add new terms to the cost function to adapt the model to specific tasks, such as clustering, or to preserve some structural properties in the reduced space (e.g., local invariance). The added term is mainly weighted by a hyper-parameter to control the balance of the overall formula to guide the optimization process towards the objective. The result is a parameterized NMF method. However, NMF method adopts unsupervised approaches to estimate the factorizing matrices. Thus, the ability to perform prediction (e.g. classification) using the new obtained features is not guaranteed. The objective of this work is to design an evolutionary framework to learn the hyper-parameter of the parameterized NMF and estimate the factorizing matrices in a supervised way to be more suitable for classification problems. Moreover, we claim that applying NMF-based algorithms separately to different class-pairs instead of applying it once to the whole dataset improves the effectiveness of the matrix factorization process. This results in training multiple parameterized NMF algorithms with different balancing parameter values. A cross-validation combination learning framework is adopted and a Genetic Algorithm is used to identify the optimal set of hyper-parameter values. The experiments we conducted on both real and synthetic datasets demonstrated the effectiveness of the proposed approach.

Accurate estimation of Intrinsic Dimensionality (ID) is of crucial importance in many data mining and machine learning tasks, including dimensionality reduction, outlier detection, similarity search and subspace clustering. However, since their convergence generally requires sample sizes (that is, neighborhood sizes) on the order of hundreds of points, existing ID estimation methods may have only limited usefulness for applications in which the data consists of many natural groups of small size. In this paper, we propose a local ID estimation strategy stable even for `tight' localities consisting of as few as 20 sample points. The estimator applies MLE techniques over all available pairwise distances among the members of the sample, based on a recent extreme-value-theoretic model of intrinsic dimensionality, the Local Intrinsic Dimension (LID). Our experimental results show that our proposed estimation technique can achieve notably smaller variance, while maintaining comparable levels of bias, at much smaller sample sizes than state-of-the-art estimators.

In contrast to multi-label learning, label distribution learning characterizes the polysemy of examples by a label distribution to represent richer semantics. In the learning process of label distribution, the training data is collected mainly by manual annotation or label enhancement algorithms to generate label distribution. Unfortunately, the complexity of the manual annotation task or the inaccuracy of the label enhancement algorithm leads to noise and uncertainty in the label distribution training set. To alleviate this problem, we introduce the implicit distribution in the label distribution learning framework to characterize the uncertainty of each label value. Specifically, we use deep implicit representation learning to construct a label distribution matrix with Gaussian prior constraints, where each row component corresponds to the distribution estimate of each label value, and this row component is constrained by a prior Gaussian distribution to moderate the noise and uncertainty interference of the label distribution dataset. Finally, each row component of the label distribution matrix is transformed into a standard label distribution form by using the self-attention algorithm. In addition, some approaches with regularization characteristics are conducted in the training phase to improve the performance of the model.

Text detection in natural scene images has applications for autonomous driving, navigation help for elderly and blind people. However, the research on Urdu text detection is usually hindered by lack of data resources. We have developed a dataset of scene images with Urdu text. We present the use of machine learning methods to perform detection of Urdu text from the scene images. We extract text regions using channel enhanced Maximally Stable Extremal Region (MSER) method. First, we classify text and noise based on their geometric properties. Next, we use a support vector machine for early discarding of non-text regions. To further remove the non-text regions, we use histogram of oriented gradients (HoG) features obtained and train a second SVM classifier. This improves the overall performance on text region detection within the scene images. To support research on Urdu text, We aim to make the data freely available for research use. We also aim to highlight the challenges and the research gap for Urdu text detection.

We consider the problem of learning the optimal threshold policy for control problems. Threshold policies make control decisions by evaluating whether an element of the system state exceeds a certain threshold, whose value is determined by other elements of the system state. By leveraging the monotone property of threshold policies, we prove that their policy gradients have a surprisingly simple expression. We use this simple expression to build an off-policy actor-critic algorithm for learning the optimal threshold policy. Simulation results show that our policy significantly outperforms other reinforcement learning algorithms due to its ability to exploit the monotone property. In addition, we show that the Whittle index, a powerful tool for restless multi-armed bandit problems, is equivalent to the optimal threshold policy for an alternative problem. This observation leads to a simple algorithm that finds the Whittle index by learning the optimal threshold policy in the alternative problem. Simulation results show that our algorithm learns the Whittle index much faster than several recent studies that learn the Whittle index through indirect means.

SINGAPORE, September 27, 2022 – Asia/Pacific* spending on AI systems (including hardware, software, and services) will rise from $20.6 billion in 2022 to around $46.6 billion in 2026, according to IDC's latest Worldwide Artificial Intelligence Spending Guide. Artificial intelligence (AI) solutions have become an essential part of process improvement and business decision-making, assisting organizations in keeping up with market viability. IDC forecasts a compound annual growth rate (CAGR) of 23.7 percent for 2021-2026. "Pre-trained natural language and computer vision models have contributed largely to the 1st wave of adoptions. It's time for more organizations to tap into their own data asset and start managing the "data to intelligence" lifecycle. This will become one of the differentiating capabilities for companies to compete in the digital-first era," says Jessie Danqing Cai, Associate Research Director, Artificial Intelligence, IDC Asia/Pacific.