Kunal Dahiya

We develop accurate and efficient solutions for large-scale retrieval tasks where novel (zero-shot) items can arrive continuously at a rapid pace. Conventional Siamese-style approaches embed both queries and items through a small encoder and retrieve the items lying closest to the query. While this approach allows efficient addition and retrieval of novel items, the small encoder lacks sufficient capacity for the necessary world knowledge in complex retrieval tasks. The extreme classification approaches have addressed this by learning a separate classifier for each item observed in the training set which significantly increases the representation capacity of the model. Such classifiers outperform Siamese approaches on observed items, but cannot be trained for novel items due to data and latency constraints. To bridge these gaps, this paper develops: (1) A new algorithmic framework, EMMETT, which efficiently synthesizes classifiers on-the-fly for novel items, by relying on the readily available classifiers for observed items; (2) A new algorithm, IRENE, which is a simple and effective instance of EMMETT that is specifically suited for large-scale deployments, and (3) A new theoretical framework for analyzing the generalization performance in large-scale zero-shot retrieval which guides our algorithm and training related design decisions. Comprehensive experiments are conducted on a wide range of retrieval tasks which demonstrate that IRENE improves the zero-shot retrieval accuracy by up to 15% points in Recall@10 when added on top of leading encoders. Additionally, on an online A/B test in a large-scale ad retrieval task in a major search engine, IRENE improved the ad click-through rate by 4.2%. Lastly, we validate our design choices through extensive ablative experiments. The source code for IRENE is available at https://aka.ms/irene.

The task of annotating a data point with labels most relevant to it from a large universe of labels is referred to as Extreme Classification (XC). State-of-the-art XC methods have applications in ranking, recommendation, and tagging and mostly employ a combination architecture comprised of a deep encoder and a high-capacity classifier. These two components are often trained in a modular fashion to conserve compute. This paper shows that in XC settings where data paucity and semantic gap issues abound, this can lead to suboptimal encoder training which negatively affects the performance of the overall architecture. The paper then proposes a lightweight alternative DEXA that augments encoder training with auxiliary parameters. Incorporating DEXA into existing XC architectures requires minimal modifications and the method can scale to datasets with 40 million labels and offer predictions that are up to 6% and 15% more accurate than embeddings offered by existing deep XC methods on benchmark and proprietary datasets, respectively. The paper also analyzes DEXA theoretically and shows that it offers provably superior encoder training than existing Siamese training strategies in certain realizable settings. Code for DEXA is available at https://github.com/Extreme-classification/dexa.

Extreme Classification (XC) seeks to tag data points with the most relevant subset of labels from an extremely large label set. Performing deep XC with dense, learnt representations for data points and labels has attracted much attention due to its superiority over earlier XC methods that used sparse, hand-crafted features. Negative mining techniques have emerged as a critical component of all deep XC methods, allowing them to scale to millions of labels. However, despite recent advances, training deep XC models with large encoder architectures such as transformers remains challenging. This paper notices that memory overheads of popular negative mining techniques often force mini-batch sizes to remain small and slow training down. In response, this paper introduces NGAME, a light-weight mini-batch creation technique that offers provably accurate in-batch negative samples. This allows training with larger mini-batches offering significantly faster convergence and higher accuracies than existing negative sampling techniques. NGAME was found to be up to 16% more accurate than state-of-the-art methods on a wide array of benchmark datasets for extreme classification, as well as 3% more accurate at retrieving search engine queries in response to a user webpage visit to show personalized ads. In live A/B tests on a popular search engine, NGAME yielded up to 23% gains in click-through-rates. Code for NGAME is available at https://github.com/Extreme-classification/ngame

Indoor thermal comfort in smart buildings has a significant impact on the health and performance of occupants. Consequently, machine learning (ML) is increasingly used to solve challenges related to indoor thermal comfort. Temporal variability of thermal comfort perception is an important problem that regulates occupant well-being and energy consumption. However, in most ML-based thermal comfort studies, temporal aspects such as the time of day, circadian rhythm, and outdoor temperature are not considered. This work addresses these problems. It investigates the impact of circadian rhythm and outdoor temperature on the prediction accuracy and classification performance of ML models. The data is gathered through month-long field experiments carried out in 14 classrooms of 5 schools, involving 512 primary school students. Four thermal comfort metrics are considered as the outputs of Deep Neural Networks and Support Vector Machine models for the dataset. The effect of temporal variability on school children's comfort is shown through a "time of day" analysis. Temporal variability in prediction accuracy is demonstrated (up to 80%). Furthermore, we show that outdoor temperature (varying over time) positively impacts the prediction performance of thermal comfort models by up to 30%. The importance of spatio-temporal context is demonstrated by contrasting micro-level (location specific) and macro-level (6 locations across a city) performance. The most important finding of this work is that a definitive improvement in prediction accuracy is shown with an increase in the time of day and sky illuminance, for multiple thermal comfort metrics.

Thermal comfort in indoor environments has an enormous impact on the health, well-being, and performance of occupants. Given the focus on energy efficiency and Internet-of-Things enabled smart buildings, machine learning (ML) is being increasingly used for data-driven thermal comfort (TC) prediction. Generally, ML-based solutions are proposed for air-conditioned or HVAC ventilated buildings and the models are primarily designed for adults. On the other hand, naturally ventilated (NV) buildings are the norm in most countries. They are also ideal for energy conservation and long-term sustainability goals. However, the indoor environment of NV buildings lacks thermal regulation and varies significantly across spatial contexts. These factors make TC prediction extremely challenging. Thus, determining the impact of the building environment on the performance of TC models is important. Further, the generalization capability of TC prediction models across different NV indoor spaces needs to be studied. This work addresses these problems. Data is gathered through month-long field experiments conducted in 5 naturally ventilated school buildings, involving 512 primary school students. The impact of spatial variability on student comfort is demonstrated through variation in prediction accuracy (by as much as 71%). The influence of building environment on TC prediction is also demonstrated through variation in feature importance. Further, a comparative analysis of spatial variability in model performance is done for children (our dataset) and adults (ASHRAE-II database). Finally, the generalization capability of thermal comfort models in NV classrooms is assessed and major challenges are highlighted.