Sajad Darabi

Sajad Darabi, Piotr Bigaj, Dawid Majchrowski et al.

Recently there has been increasing interest in developing and deploying deep graph learning algorithms for many tasks, such as fraud detection and recommender systems. Albeit, there is a limited number of publicly available graph-structured datasets, most of which are tiny compared to production-sized applications or are limited in their application domain. This work tackles this shortcoming by proposing a scalable synthetic graph generation tool to scale the datasets to production-size graphs with trillions of edges and billions of nodes. The tool learns a series of parametric models from proprietary datasets that can be released to researchers to study various graph methods on the synthetic data increasing prototype development and novel applications. We demonstrate the generalizability of the framework across a series of datasets, mimicking structural and feature distributions as well as the ability to scale them across varying sizes demonstrating their usefulness for benchmarking and model development. Code can be found on https://github.com/NVIDIA/DeepLearningExamples/tree/master/Tools/DGLPyTorch/SyntheticGraphGeneration.

Sajad Darabi, Shayan Fazeli, Jiwei Liu et al.

Previous works have demonstrated the importance of considering different modalities on molecules, each of which provide a varied granularity of information for downstream property prediction tasks. Our method combines variants of the recent TransformerM architecture with Transformer, GNN, and ResNet backbone architectures. Models are trained on the 2D data, 3D data, and image modalities of molecular graphs. We ensemble these models with a HuberRegressor. The models are trained on 4 different train/validation splits of the original train + valid datasets. This yields a winning solution to the 2\textsuperscript{nd} edition of the OGB Large-Scale Challenge (2022) on the PCQM4Mv2 molecular property prediction dataset. Our proposed method achieves a test-challenge MAE of $0.0723$ and a validation MAE of $0.07145$. Total inference time for our solution is less than 2 hours. We open-source our code at https://github.com/jfpuget/NVIDIA-PCQM4Mv2.

Peter St. John, Dejun Lin, Polina Binder et al.

Artificial Intelligence models encoding biology and chemistry are opening new routes to high-throughput and high-quality in-silico drug development. However, their training increasingly relies on computational scale, with recent protein language models (pLM) training on hundreds of graphical processing units (GPUs). We introduce the BioNeMo Framework to facilitate the training of computational biology and chemistry AI models across hundreds of GPUs. Its modular design allows the integration of individual components, such as data loaders, into existing workflows and is open to community contributions. We detail technical features of the BioNeMo Framework through use cases such as pLM pre-training and fine-tuning. On 256 NVIDIA A100s, BioNeMo Framework trains a three billion parameter BERT-based pLM on over one trillion tokens in 4.2 days. The BioNeMo Framework is open-source and free for everyone to use.

Sajad Darabi, Mohammad Kachuee, Shayan Fazeli et al.

Effective representation learning of electronic health records is a challenging task and is becoming more important as the availability of such data is becoming pervasive. The data contained in these records are irregular and contain multiple modalities such as notes, and medical codes. They are preempted by medical conditions the patient may have, and are typically jotted down by medical staff. Accompanying codes are notes containing valuable information about patients beyond the structured information contained in electronic health records. We use transformer networks and the recently proposed BERT language model to embed these data streams into a unified vector representation. The presented approach effectively encodes a patient's visit data into a single distributed representation, which can be used for downstream tasks. Our model demonstrates superior performance and generalization on mortality, readmission and length of stay tasks using the publicly available MIMIC-III ICU dataset. Code avaialble at https://github.com/sajaddarabi/TAPER-EHR

Jan Bączek, Dmytro Zhylko, Gilberto Titericz et al.

While machine learning has witnessed significant advancements, the emphasis has largely been on data acquisition and model creation. However, achieving a comprehensive assessment of machine learning solutions in real-world settings necessitates standardization throughout the entire pipeline. This need is particularly acute in time series forecasting, where diverse settings impede meaningful comparisons between various methods. To bridge this gap, we propose a unified benchmarking framework that exposes the crucial modelling and machine learning decisions involved in developing time series forecasting models. This framework fosters seamless integration of models and datasets, aiding both practitioners and researchers in their development efforts. We benchmark recently proposed models within this framework, demonstrating that carefully implemented deep learning models with minimal effort can rival gradient-boosting decision trees requiring extensive feature engineering and expert knowledge.

Sajad Darabi, Shayan Fazeli, Ali Pazoki et al.

Recent literature in self-supervised has demonstrated significant progress in closing the gap between supervised and unsupervised methods in the image and text domains. These methods rely on domain-specific augmentations that are not directly amenable to the tabular domain. Instead, we introduce Contrastive Mixup, a semi-supervised learning framework for tabular data and demonstrate its effectiveness in limited annotated data settings. Our proposed method leverages Mixup-based augmentation under the manifold assumption by mapping samples to a low dimensional latent space and encourage interpolated samples to have high a similarity within the same labeled class. Unlabeled samples are additionally employed via a transductive label propagation method to further enrich the set of similar and dissimilar pairs that can be used in the contrastive loss term. We demonstrate the effectiveness of the proposed framework on public tabular datasets and real-world clinical datasets.

Sajad Darabi, Yotam Elor

Real-world binary classification tasks are in many cases imbalanced, where the minority class is much smaller than the majority class. This skewness is challenging for machine learning algorithms as they tend to focus on the majority and greatly misclassify the minority. Adding synthetic minority samples to the dataset before training the model is a popular technique to address this difficulty and is commonly achieved by interpolating minority samples. Tabular datasets are often multi-modal and contain discrete (categorical) features in addition to continuous ones which makes interpolation of samples non-trivial. To address this, we propose a latent space interpolation framework which (1) maps the multi-modal samples to a dense continuous latent space using an autoencoder; (2) applies oversampling by interpolation in the latent space; and (3) maps the synthetic samples back to the original feature space. We defined metrics to directly evaluate the quality of the minority data generated and showed that our framework generates better synthetic data than the existing methods. Furthermore, the superior synthetic data yields better prediction quality in downstream binary classification tasks, as was demonstrated in extensive experiments with 27 publicly available real-world datasets

Mohammad Kachuee, Sajad Darabi, Shayan Fazeli et al.

Despite the success of deep learning in domains such as image, voice, and graphs, there has been little progress in deep representation learning for domains without a known structure between features. For instance, a tabular dataset of different demographic and clinical factors where the feature interactions are not given as a prior. In this paper, we propose Group-Connected Multilayer Perceptron (GMLP) networks to enable deep representation learning in these domains. GMLP is based on the idea of learning expressive feature combinations (groups) and exploiting them to reduce the network complexity by defining local group-wise operations. During the training phase, GMLP learns a sparse feature grouping matrix using temperature annealing softmax with an added entropy loss term to encourage the sparsity. Furthermore, an architecture is suggested which resembles binary trees, where group-wise operations are followed by pooling operations to combine information; reducing the number of groups as the network grows in depth. To evaluate the proposed method, we conducted experiments on different real-world datasets covering various application areas. Additionally, we provide visualizations on MNIST and synthesized data. According to the results, GMLP is able to successfully learn and exploit expressive feature combinations and achieve state-of-the-art classification performance on different datasets.

Sajad Darabi, Mohammad Kachuee, Majid Sarrafzadeh

Effective modeling of electronic health records presents many challenges as they contain large amounts of irregularity most of which are due to the varying procedures and diagnosis a patient may have. Despite the recent progress in machine learning, unsupervised learning remains largely at open, especially in the healthcare domain. In this work, we present a two-step unsupervised representation learning scheme to summarize the multi-modal clinical time series consisting of signals and medical codes into a patient status vector. First, an auto-encoder step is used to reduce sparse medical codes and clinical time series into a distributed representation. Subsequently, the concatenation of the distributed representations is further fine-tuned using a forecasting task. We evaluate the usefulness of the representation on two downstream tasks: mortality and readmission. Our proposed method shows improved generalization performance for both short duration ICU visits and long duration ICU visits.

Mohammad Kachuee, Kimmo Karkkainen, Orpaz Goldstein et al.

In many machine learning applications, we are faced with incomplete datasets. In the literature, missing data imputation techniques have been mostly concerned with filling missing values. However, the existence of missing values is synonymous with uncertainties not only over the distribution of missing values but also over target class assignments that require careful consideration. In this paper, we propose a simple and effective method for imputing missing features and estimating the distribution of target assignments given incomplete data. In order to make imputations, we train a simple and effective generator network to generate imputations that a discriminator network is tasked to distinguish. Following this, a predictor network is trained using the imputed samples from the generator network to capture the classification uncertainties and make predictions accordingly. The proposed method is evaluated on CIFAR-10 and MNIST image datasets as well as five real-world tabular classification datasets, under different missingness rates and structures. Our experimental results show the effectiveness of the proposed method in generating imputations as well as providing estimates for the class uncertainties in a classification task when faced with missing values.

Mouloud Belbahri, Eyyüb Sari, Sajad Darabi et al.

Deep neural networks (DNNs) have demonstrated success for many supervised learning tasks, ranging from voice recognition, object detection, to image classification. However, their increasing complexity might yield poor generalization error that make them hard to be deployed on edge devices. Quantization is an effective approach to compress DNNs in order to meet these constraints. Using a quasiconvex base function in order to construct a binary quantizer helps training binary neural networks (BNNs) and adding noise to the input data or using a concrete regularization function helps to improve generalization error. Here we introduce foothill function, an infinitely differentiable quasiconvex function. This regularizer is flexible enough to deform towards $L_1$ and $L_2$ penalties. Foothill can be used as a binary quantizer, as a regularizer, or as a loss. In particular, we show this regularizer reduces the accuracy gap between BNNs and their full-precision counterpart for image classification on ImageNet.

Mohammad Kachuee, Orpaz Goldstein, Kimmo Karkkainen et al.

In many real-world learning scenarios, features are only acquirable at a cost constrained under a budget. In this paper, we propose a novel approach for cost-sensitive feature acquisition at the prediction-time. The suggested method acquires features incrementally based on a context-aware feature-value function. We formulate the problem in the reinforcement learning paradigm, and introduce a reward function based on the utility of each feature. Specifically, MC dropout sampling is used to measure expected variations of the model uncertainty which is used as a feature-value function. Furthermore, we suggest sharing representations between the class predictor and value function estimator networks. The suggested approach is completely online and is readily applicable to stream learning setups. The solution is evaluated on three different datasets including the well-known MNIST dataset as a benchmark as well as two cost-sensitive datasets: Yahoo Learning to Rank and a dataset in the medical domain for diabetes classification. According to the results, the proposed method is able to efficiently acquire features and make accurate predictions.

Sajad Darabi, Mouloud Belbahri, Matthieu Courbariaux et al.

There is a significant performance gap between Binary Neural Networks (BNNs) and floating point Deep Neural Networks (DNNs). We propose to improve the binary training method, by introducing a new regularization function that encourages training weights around binary values. In addition, we add trainable scaling factors to our regularization functions. Additionally, an improved approximation of the derivative of the sign activation function in the backward computation. These modifications are based on linear operations that are easily implementable into the binary training framework. Experimental results on ImageNet shows our method outperforms the traditional BNN method and XNOR-net.

Mohammad Kachuee, Sajad Darabi, Babak Moatamed et al.

In real-world scenarios, different features have different acquisition costs at test-time which necessitates cost-aware methods to optimize the cost and performance trade-off. This paper introduces a novel and scalable approach for cost-aware feature acquisition at test-time. The method incrementally asks for features based on the available context that are known feature values. The proposed method is based on sensitivity analysis in neural networks and density estimation using denoising autoencoders with binary representation layers. In the proposed architecture, a denoising autoencoder is used to handle unknown features (i.e., features that are yet to be acquired), and the sensitivity of predictions with respect to each unknown feature is used as a context-dependent measure of informativeness. We evaluated the proposed method on eight different real-world datasets as well as one synthesized dataset and compared its performance with several other approaches in the literature. According to the results, the suggested method is capable of efficiently acquiring features at test-time in a cost- and context-aware fashion.