Abubakar Abid

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Chris Chinenye Emezue, Sanchit Gandhi, Lewis Tunstall et al. · huggingface

The advancement of speech technologies has been remarkable, yet its integration with African languages remains limited due to the scarcity of African speech corpora. To address this issue, we present AfroDigits, a minimalist, community-driven dataset of spoken digits for African languages, currently covering 38 African languages. As a demonstration of the practical applications of AfroDigits, we conduct audio digit classification experiments on six African languages [Igbo (ibo), Yoruba (yor), Rundi (run), Oshiwambo (kua), Shona (sna), and Oromo (gax)] using the Wav2Vec2.0-Large and XLS-R models. Our experiments reveal a useful insight on the effect of mixing African speech corpora during finetuning. AfroDigits is the first published audio digit dataset for African languages and we believe it will, among other things, pave the way for Afro-centric speech applications such as the recognition of telephone numbers, and street numbers. We release the dataset and platform publicly at https://huggingface.co/datasets/chrisjay/crowd-speech-africa and https://huggingface.co/spaces/chrisjay/afro-speech respectively.

Ammar Sherif, Abubakar Abid, Mustafa Elattar et al.

Multi-Task Learning (MTL) is a powerful technique that has gained popularity due to its performance improvement over traditional Single-Task Learning (STL). However, MTL is often challenging because there is an exponential number of possible task groupings, which can make it difficult to choose the best one because some groupings might produce performance degradation due to negative interference between tasks. That is why existing solutions are severely suffering from scalability issues, limiting any practical application. In our paper, we propose a new data-driven method that addresses these challenges and provides a scalable and modular solution for classification task grouping based on a re-proposed data-driven features, Data Maps, which capture the training dynamics for each classification task during the MTL training. Through a theoretical comparison with other techniques, we manage to show that our approach has the superior scalability. Our experiments show a better performance and verify the method's effectiveness, even on an unprecedented number of tasks (up to 100 tasks on CIFAR100). Being the first to work on such number of tasks, our comparisons on the resulting grouping shows similar grouping to the mentioned in the dataset, CIFAR100. Finally, we provide a modular implementation for easier integration and testing, with examples from multiple datasets and tasks.

Kailas Vodrahalli, Justin Ko, Albert S. Chiou et al.

Telemedicine utilization was accelerated during the COVID-19 pandemic, and skin conditions were a common use case. However, the quality of photographs sent by patients remains a major limitation. To address this issue, we developed TrueImage 2.0, an artificial intelligence (AI) model for assessing patient photo quality for telemedicine and providing real-time feedback to patients for photo quality improvement. TrueImage 2.0 was trained on 1700 telemedicine images annotated by clinicians for photo quality. On a retrospective dataset of 357 telemedicine images, TrueImage 2.0 effectively identified poor quality images (Receiver operator curve area under the curve (ROC-AUC) =0.78) and the reason for poor quality (Blurry ROC-AUC=0.84, Lighting issues ROC-AUC=0.70). The performance is consistent across age, gender, and skin tone. Next, we assessed whether patient-TrueImage 2.0 interaction led to an improvement in submitted photo quality through a prospective clinical pilot study with 98 patients. TrueImage 2.0 reduced the number of patients with a poor-quality image by 68.0%.

Abubakar Abid, Ali Abdalla, Ali Abid et al.

Accessibility is a major challenge of machine learning (ML). Typical ML models are built by specialists and require specialized hardware/software as well as ML experience to validate. This makes it challenging for non-technical collaborators and endpoint users (e.g. physicians) to easily provide feedback on model development and to gain trust in ML. The accessibility challenge also makes collaboration more difficult and limits the ML researcher's exposure to realistic data and scenarios that occur in the wild. To improve accessibility and facilitate collaboration, we developed an open-source Python package, Gradio, which allows researchers to rapidly generate a visual interface for their ML models. Gradio makes accessing any ML model as easy as sharing a URL. Our development of Gradio is informed by interviews with a number of machine learning researchers who participate in interdisciplinary collaborations. Their feedback identified that Gradio should support a variety of interfaces and frameworks, allow for easy sharing of the interface, allow for input manipulation and interactive inference by the domain expert, as well as allow embedding the interface in iPython notebooks. We developed these features and carried out a case study to understand Gradio's usefulness and usability in the setting of a machine learning collaboration between a researcher and a cardiologist.

Tarek Naous, Srinjay Sarkar, Abubakar Abid et al.

Clustering algorithms are one of the main analytical methods to detect patterns in unlabeled data. Existing clustering methods typically treat samples in a dataset as points in a metric space and compute distances to group together similar points. In this paper, we present a wholly different way of clustering points in 2-dimensional space, inspired by how humans cluster data: by training neural networks to perform instance segmentation on plotted data. Our approach, Visual Clustering, has several advantages over traditional clustering algorithms: it is much faster than most existing clustering algorithms (making it suitable for very large datasets), it agrees strongly with human intuition for clusters, and it is by default hyperparameter free (although additional steps with hyperparameters can be introduced for more control of the algorithm). We describe the method and compare it to ten other clustering methods on synthetic data to illustrate its advantages and disadvantages. We then demonstrate how our approach can be extended to higher dimensional data and illustrate its performance on real-world data. The implementation of Visual Clustering is publicly available and can be applied to any dataset in a few lines of code.

Abubakar Abid, Mert Yuksekgonul, James Zou

Understanding and explaining the mistakes made by trained models is critical to many machine learning objectives, such as improving robustness, addressing concept drift, and mitigating biases. However, this is often an ad hoc process that involves manually looking at the model's mistakes on many test samples and guessing at the underlying reasons for those incorrect predictions. In this paper, we propose a systematic approach, conceptual counterfactual explanations (CCE), that explains why a classifier makes a mistake on a particular test sample(s) in terms of human-understandable concepts (e.g. this zebra is misclassified as a dog because of faint stripes). We base CCE on two prior ideas: counterfactual explanations and concept activation vectors, and validate our approach on well-known pretrained models, showing that it explains the models' mistakes meaningfully. In addition, for new models trained on data with spurious correlations, CCE accurately identifies the spurious correlation as the cause of model mistakes from a single misclassified test sample. On two challenging medical applications, CCE generated useful insights, confirmed by clinicians, into biases and mistakes the model makes in real-world settings.

Abubakar Abid, Maheen Farooqi, James Zou

It has been observed that large-scale language models capture undesirable societal biases, e.g. relating to race and gender; yet religious bias has been relatively unexplored. We demonstrate that GPT-3, a state-of-the-art contextual language model, captures persistent Muslim-violence bias. We probe GPT-3 in various ways, including prompt completion, analogical reasoning, and story generation, to understand this anti-Muslim bias, demonstrating that it appears consistently and creatively in different uses of the model and that it is severe even compared to biases about other religious groups. For instance, "Muslim" is analogized to "terrorist" in 23% of test cases, while "Jewish" is mapped to "money" in 5% of test cases. We quantify the positive distraction needed to overcome this bias with adversarial text prompts, and find that use of the most positive 6 adjectives reduces violent completions for "Muslims" from 66% to 20%, but which is still higher than for other religious groups.

Kexin Huang, Tianfan Fu, Dawood Khan et al.

The efficacy of a drug depends on its binding affinity to the therapeutic target and pharmacokinetics. Deep learning (DL) has demonstrated remarkable progress in predicting drug efficacy. We develop MolDesigner, a human-in-the-loop web user-interface (UI), to assist drug developers leverage DL predictions to design more effective drugs. A developer can draw a drug molecule in the interface. In the backend, more than 17 state-of-the-art DL models generate predictions on important indices that are crucial for a drug's efficacy. Based on these predictions, drug developers can edit the drug molecule and reiterate until satisfaction. MolDesigner can make predictions in real-time with a latency of less than a second.

Abubakar Abid, James Zou

Fine-grained annotations---e.g. dense image labels, image segmentation and text tagging---are useful in many ML applications but they are labor-intensive to generate. Moreover there are often systematic, structured errors in these fine-grained annotations. For example, a car might be entirely unannotated in the image, or the boundary between a car and street might only be coarsely annotated. Standard ML training on data with such structured errors produces models with biases and poor performance. In this work, we propose a novel framework of Error-Correcting Networks (ECN) to address the challenge of learning in the presence structured error in fine-grained annotations. Given a large noisy dataset with commonly occurring structured errors, and a much smaller dataset with more accurate annotations, ECN is able to substantially improve the prediction of fine-grained annotations compared to standard approaches for training on noisy data. It does so by learning to leverage the structures in the annotations and in the noisy labels. Systematic experiments on image segmentation and text tagging demonstrate the strong performance of ECN in improving training on noisy structured labels.

Abubakar Abid, James Zou

Variational autoencoders are powerful algorithms for identifying dominant latent structure in a single dataset. In many applications, however, we are interested in modeling latent structure and variation that are enriched in a target dataset compared to some background---e.g. enriched in patients compared to the general population. Contrastive learning is a principled framework to capture such enriched variation between the target and background, but state-of-the-art contrastive methods are limited to linear models. In this paper, we introduce the contrastive variational autoencoder (cVAE), which combines the benefits of contrastive learning with the power of deep generative models. The cVAE is designed to identify and enhance salient latent features. The cVAE is trained on two related but unpaired datasets, one of which has minimal contribution from the salient latent features. The cVAE explicitly models latent features that are shared between the datasets, as well as those that are enriched in one dataset relative to the other, which allows the algorithm to isolate and enhance the salient latent features. The algorithm is straightforward to implement, has a similar run-time to the standard VAE, and is robust to noise and dataset purity. We conduct experiments across diverse types of data, including gene expression and facial images, showing that the cVAE effectively uncovers latent structure that is salient in a particular analysis.

Abubakar Abid, Muhammad Fatih Balin, James Zou

We introduce the concrete autoencoder, an end-to-end differentiable method for global feature selection, which efficiently identifies a subset of the most informative features and simultaneously learns a neural network to reconstruct the input data from the selected features. Our method is unsupervised, and is based on using a concrete selector layer as the encoder and using a standard neural network as the decoder. During the training phase, the temperature of the concrete selector layer is gradually decreased, which encourages a user-specified number of discrete features to be learned. During test time, the selected features can be used with the decoder network to reconstruct the remaining input features. We evaluate concrete autoencoders on a variety of datasets, where they significantly outperform state-of-the-art methods for feature selection and data reconstruction. In particular, on a large-scale gene expression dataset, the concrete autoencoder selects a small subset of genes whose expression levels can be use to impute the expression levels of the remaining genes. In doing so, it improves on the current widely-used expert-curated L1000 landmark genes, potentially reducing measurement costs by 20%. The concrete autoencoder can be implemented by adding just a few lines of code to a standard autoencoder.

Abdi-Hakin Dirie, Abubakar Abid, James Zou

We introduce Contrastive Multivariate Singular Spectrum Analysis, a novel unsupervised method for dimensionality reduction and signal decomposition of time series data. By utilizing an appropriate background dataset, the method transforms a target time series dataset in a way that evinces the sub-signals that are enhanced in the target dataset, as opposed to only those that account for the greatest variance. This shifts the goal from finding signals that explain the most variance to signals that matter the most to the analyst. We demonstrate our method on an illustrative synthetic example, as well as show the utility of our method in the downstream clustering of electrocardiogram signals from the public MHEALTH dataset.

Abubakar Abid, James Zou

Measuring similarities between unlabeled time series trajectories is an important problem in domains as diverse as medicine, astronomy, finance, and computer vision. It is often unclear what is the appropriate metric to use because of the complex nature of noise in the trajectories (e.g. different sampling rates or outliers). Domain experts typically hand-craft or manually select a specific metric, such as dynamic time warping (DTW), to apply on their data. In this paper, we propose Autowarp, an end-to-end algorithm that optimizes and learns a good metric given unlabeled trajectories. We define a flexible and differentiable family of warping metrics, which encompasses common metrics such as DTW, Euclidean, and edit distance. Autowarp then leverages the representation power of sequence autoencoders to optimize for a member of this warping distance family. The output is a metric which is easy to interpret and can be robustly learned from relatively few trajectories. In systematic experiments across different domains, we show that Autowarp often outperforms hand-crafted trajectory similarity metrics.

Abubakar Abid, James Zou

We consider the problem of inference in a linear regression model in which the relative ordering of the input features and output labels is not known. Such datasets naturally arise from experiments in which the samples are shuffled or permuted during the protocol. In this work, we propose a framework that treats the unknown permutation as a latent variable. We maximize the likelihood of observations using a stochastic expectation-maximization (EM) approach. We compare this to the dominant approach in the literature, which corresponds to hard EM in our framework. We show on synthetic data that the stochastic EM algorithm we develop has several advantages, including lower parameter error, less sensitivity to the choice of initialization, and significantly better performance on datasets that are only partially shuffled. We conclude by performing two experiments on real datasets that have been partially shuffled, in which we show that the stochastic EM algorithm can recover the weights with modest error.

Amirata Ghorbani, Abubakar Abid, James Zou

In order for machine learning to be deployed and trusted in many applications, it is crucial to be able to reliably explain why the machine learning algorithm makes certain predictions. For example, if an algorithm classifies a given pathology image to be a malignant tumor, then the doctor may need to know which parts of the image led the algorithm to this classification. How to interpret black-box predictors is thus an important and active area of research. A fundamental question is: how much can we trust the interpretation itself? In this paper, we show that interpretation of deep learning predictions is extremely fragile in the following sense: two perceptively indistinguishable inputs with the same predicted label can be assigned very different interpretations. We systematically characterize the fragility of several widely-used feature-importance interpretation methods (saliency maps, relevance propagation, and DeepLIFT) on ImageNet and CIFAR-10. Our experiments show that even small random perturbation can change the feature importance and new systematic perturbations can lead to dramatically different interpretations without changing the label. We extend these results to show that interpretations based on exemplars (e.g. influence functions) are similarly fragile. Our analysis of the geometry of the Hessian matrix gives insight on why fragility could be a fundamental challenge to the current interpretation approaches.

Abubakar Abid, Martin J. Zhang, Vivek K. Bagaria et al.

We present a new technique called contrastive principal component analysis (cPCA) that is designed to discover low-dimensional structure that is unique to a dataset, or enriched in one dataset relative to other data. The technique is a generalization of standard PCA, for the setting where multiple datasets are available -- e.g. a treatment and a control group, or a mixed versus a homogeneous population -- and the goal is to explore patterns that are specific to one of the datasets. We conduct a wide variety of experiments in which cPCA identifies important dataset-specific patterns that are missed by PCA, demonstrating that it is useful for many applications: subgroup discovery, visualizing trends, feature selection, denoising, and data-dependent standardization. We provide geometrical interpretations of cPCA and show that it satisfies desirable theoretical guarantees. We also extend cPCA to nonlinear settings in the form of kernel cPCA. We have released our code as a python package and documentation is on Github.

Abubakar Abid, Ada Poon, James Zou

Is it possible to perform linear regression on datasets whose labels are shuffled with respect to the inputs? We explore this question by proposing several estimators that recover the weights of a noisy linear model from labels that are shuffled by an unknown permutation. We show that the analog of the classical least-squares estimator produces inconsistent estimates in this setting, and introduce an estimator based on the self-moments of the input features and labels. We study the regimes in which each estimator excels, and generalize the estimators to the setting where partial ordering information is available in the form of experiments replicated independently. The result is a framework that enables robust inference, as we demonstrate by experiments on both synthetic and standard datasets, where we are able to recover approximate weights using only shuffled labels. Our work demonstrates that linear regression in the absence of complete ordering information is possible and can be of practical interest, particularly in experiments that characterize populations of particles, such as flow cytometry.