Determining the precise location of customers is important for an efficient and reliable delivery experience, both for customers and delivery associates. Address text is a primary source of information provided by customers about their location. In this paper, we study the important and challenging task of matching free-form customer address text to determine if two addresses represent the same physical building. We introduce a novel address matching framework that leverages transformer-based encoder to prevent tedious and time-consuming efforts spent on manual feature engineering by the baseline model. Furthermore, our proposed framework employs weak supervision to leverage historic delivery information and generate high-quality labeled data. This reduces the requirement for massive amounts of labeled data, typically needed for transformer-based models. Our experiments on manually curated datasets demonstrate the effective and generic nature of our approach, as we achieve 15.57% improvement in recall at 95% precision, on average, compared to the current baseline model across four geographies. We also introduce delivery point (DP) geocode learning for cold-start addresses as a downstream application of customer address matching. In addition to offline experiments, we performed online A/B experiments for DP geocode learning with our proposed approach and observed delivery precision improved by 8.09% and delivery defects reduced by 11.78% on average across four geographies in comparison to the baseline model.

We study an important and challenging task of attacking natural language processing models in a hard label black box setting. We propose a decision-based attack strategy that crafts high quality adversarial examples on text classification and entailment tasks. Our proposed attack strategy leverages population-based optimization algorithm to craft plausible and semantically similar adversarial examples by observing only the top label predicted by the target model. At each iteration, the optimization procedure allow word replacements that maximizes the overall semantic similarity between the original and the adversarial text. Further, our approach does not rely on using substitute models or any kind of training data. We demonstrate the efficacy of our proposed approach through extensive experimentation and ablation studies on five state-of-the-art target models across seven benchmark datasets. In comparison to attacks proposed in prior literature, we are able to achieve a higher success rate with lower word perturbation percentage that too in a highly restricted setting.

We study an important task of attacking natural language processing models in a black box setting. We propose an attack strategy that crafts semantically similar adversarial examples on text classification and entailment tasks. Our proposed attack finds candidate words by considering the information of both the original word and its surrounding context. It jointly leverages masked language modelling and next sentence prediction for context understanding. In comparison to attacks proposed in prior literature, we are able to generate high quality adversarial examples that do significantly better both in terms of success rate and word perturbation percentage.

Existing black box search methods have achieved high success rate in generating adversarial attacks against NLP models.However, such search methods are inefficient as they do not consider the amount of queries required to generate adversarial attacks.Also, prior attacks do not maintain a consistent search space while comparing different search methods.In this paper, we propose a query efficient attack strategy to generate plausible adversarial examples on text classification and entailment tasks.Our attack jointly leverages attention mechanism and locality sensitive hashing (LSH) to reduce the query count.We demonstrate the efficacy of our approach by comparing our attack with four baselines across three different search spaces.Further, we benchmark our results across the same search space used in prior attacks.In comparison to attacks proposed, on an average, we are able to reduce the query count by 75% across all datasets and target models.We also demonstrate that our attack achieves a higher success rate when compared to prior attacks in a limited query setting.

In recent years, the importance of feature engineering has been confirmed by the exceptional performance of deep learning techniques, that automate this task for some applications. For others, feature engineering requires substantial manual effort in designing and selecting features and is often tedious and non-scalable. We present AutoLearn, a regression-based feature learning algorithm. Being data-driven, it requires no domain knowledge and is hence generic. Such a representation is learnt by mining pairwise feature associations, identifying the linear or non-linear relationship between each pair, applying regression and selecting those relationships that are stable and improve the prediction performance. Our experimental evaluation on 18 UC Irvine and 7 Gene expression datasets, across different domains, provides evidence that the features learnt through our model can improve the overall prediction accuracy by 13.28%, compared to original feature space and 5.87% over other top performing models, across 8 different classifiers without using any domain knowledge.