Self-consistency-based approaches, which involve repeatedly sampling multiple outputs and selecting the most consistent one as the final response, prove to be remarkably effective in improving the factual accuracy of large language models. Nonetheless, existing methods usually have strict constraints on the task format, largely limiting their applicability. In this paper, we present Integrative Decoding (ID), to unlock the potential of self-consistency in open-ended generation tasks. ID operates by constructing a set of inputs, each prepended with a previously sampled response, and then processes them concurrently, with the next token being selected by aggregating of all their corresponding predictions at each decoding step. In essence, this simple approach implicitly incorporates self-consistency in the decoding objective. Extensive evaluation shows that ID consistently enhances factuality over a wide range of language models, with substantial improvements on the TruthfulQA (+11.2%), Biographies (+15.4%) and LongFact (+8.5%) benchmarks. The performance gains amplify progressively as the number of sampled responses increases, indicating the potential of ID to scale up with repeated sampling.

Retrieval-augmented classification and generation models benefit from *early-stage fusion* of high-quality text-based metadata, often called memory, but face high latency and noise sensitivity. In extreme classification (XC), where low latency is crucial, existing methods use *late-stage fusion* for efficiency and robustness. To enhance accuracy while maintaining low latency, we propose MOGIC, a novel approach to metadata-infused oracle guidance for XC. We train an early-fusion oracle classifier with access to both query-side and label-side ground-truth metadata in textual form and subsequently use it to guide existing memory-based XC disciple models via regularization. The MOGIC algorithm improves precision@1 and propensity-scored precision@1 of XC disciple models by 1-2% on six standard datasets, at no additional inference-time cost. We show that MOGIC can be used in a plug-and-play manner to enhance memory-free XC models such as NGAME or DEXA. Lastly, we demonstrate the robustness of the MOGIC algorithm to missing and noisy metadata. The code is publicly available at [https://github.com/suchith720/mogic](https://github.com/suchith720/mogic).

Extreme Classification (XC) aims to map a query to the most relevant documents from a very large document set. XC algorithms used in real-world applications typically learn this mapping from datasets curated from implicit feedback, such as user clicks. However, these datasets often suffer from missing labels. In this work, we observe that systematic missing labels lead to missing knowledge, which is critical for modelling relevance between queries and documents. We formally show that this absence of knowledge is hard to recover using existing methods such as propensity weighting and data imputation strategies that solely rely on the training dataset. While Large Language Models (LLMs) provide an attractive solution to augment the missing knowledge, leveraging them in applications with low latency requirements and large document sets is challenging. To mitigate missing knowledge at scale, we propose SKIM (Scalable Knowledge Infusion for Missing Labels), an algorithm that leverages a combination of Small Language Models or SLMs, e.g., Llama2-7b, and abundant unstructured meta-data to effectively address the missing label problem. We show the efficacy of our method on large-scale public datasets through a combination of unbiased evaluation strategies, such as exhaustive human annotations and simulation-based evaluation benchmarks. SKIM outperforms existing methods on Recall@100 by more than 10 absolute points. Additionally, SKIM scales to proprietary query-ad retrieval datasets containing 10 million documents, outperforming baseline methods by 12% in offline evaluations and increasing ad click-yield by 1.23% in an online A/B test conducted on Bing Search. We release the code and trained models at: github.com/bicycleman15/skim

Generative Retrieval introduces a new approach to Information Retrieval by reframing it as a constrained generation task, leveraging recent advancements in Autoregressive (AR) language models. However, AR-based Generative Retrieval methods suffer from high inference latency and cost compared to traditional dense retrieval techniques, limiting their practical applicability. This paper investigates fully Non-autoregressive (NAR) language models as a more efficient alternative for generative retrieval. While standard NAR models alleviate latency and cost concerns, they exhibit a significant drop in retrieval performance (compared to AR models) due to their inability to capture dependencies between target tokens. To address this, we question the conventional choice of limiting the target token space to solely words or sub-words. We propose PIXNAR, a novel approach that expands the target vocabulary of NAR models to include multi-word entities and common phrases (up to 5 million tokens), thereby reducing token dependencies. PIXNAR employs inference optimization strategies to maintain low inference latency despite the significantly larger vocabulary. Our results demonstrate that PIXNAR achieves a relative improvement of 31.0% in MRR@10 on MS MARCO and 23.2% in Hits@5 on Natural Questions compared to standard NAR models with similar latency and cost. Furthermore, online A/B experiments on a large commercial search engine show significant increase in clicks and revenue.

Large Language Models (LLMs) have shown superior performance in various applications and fields.To achieve better performance on specialized domains such as law and advertisement, LLMs are often continue pre-trained on in-domain data.However, existing approaches suffer from two major issues.First, in-domain data are scarce compared with general domainagnostic data.Second, data used for continual pre-training are not task-aware, such that they may not be helpful to downstream applications.We propose TRAIT, a task-oriented in-domain data augmentation framework.Our framework is divided into two parts: in-domain data selection and task-oriented synthetic passage generation.The data selection strategy identifies and selects a large amount of in-domain data from general corpora, and thus significantly enriches domain knowledge in the continual pre-training data.The synthetic passages contain guidance on how to use domain knowledge to answer questions about downstream tasks.By training on such passages, the model aligns with the need of downstream applications.We adapt LLMs to two domains: advertisement and math.On average, TRAIT improves LLM performance by 8% in the advertisement domain and 7.5% in the math domain.

Large language models (LLMs) have made impressive progress in natural language processing. These models rely on proper human instructions (or prompts) to generate suitable responses. However, the potential of LLMs are not fully harnessed by commonly-used prompting methods: many human-in-the-loop algorithms employ ad-hoc procedures for prompt selection; while auto prompt generation approaches are essentially searching all possible prompts randomly and inefficiently. We propose Evoke, an automatic prompt refinement framework. In Evoke, there are two instances of a same LLM: one as a reviewer (LLM-Reviewer), it scores the current prompt; the other as an author (LLM-Author), it edits the prompt by considering the edit history and the reviewer's feedback. Such an author-reviewer feedback loop ensures that the prompt is refined in each iteration. We further aggregate a data selection approach to Evoke, where only the hard samples are exposed to the LLM. The hard samples are more important because the LLM can develop deeper understanding of the tasks out of them, while the model may already know how to solve the easier cases. Experimental results show that Evoke significantly outperforms existing methods. For instance, in the challenging task of logical fallacy detection, Evoke scores above 80, while all other baseline methods struggle to reach 20.

The objective in eXtreme Classification (XC) is to find relevant labels for a document from an exceptionally large label space. Most XC application scenarios have rich auxiliary data associated with the input documents, e.g., frequently clicked webpages for search queries in sponsored search. Unfortunately, most of the existing XC methods do not use any auxiliary data. In this paper, we propose a novel framework, Online Auxiliary Knowledge (OAK), which harnesses auxiliary information linked to the document to improve XC accuracy. OAK stores information learnt from the auxiliary data in a knowledge bank and during a forward pass, retrieves relevant auxiliary knowledge embeddings for a given document. An enriched embedding is obtained by fusing these auxiliary knowledge embeddings with the document's embedding, thereby enabling much more precise candidate label selection and final classification. OAK training involves three stages. (1) Training a linker module to link documents to relevant auxiliary data points. (2) Learning an embedding for documents enriched using linked auxiliary information. (3) Using the enriched document embeddings to learn the final classifiers. OAK outperforms current state-of-the-art XC methods by up to $\sim 5 \%$ on academic datasets, and by $\sim 3 \%$ on an auxiliary data-augmented variant of LF-ORCAS-800K dataset in Precision@1. OAK also demonstrates statistically significant improvements in sponsored search metrics when deployed on a large scale search engine.

Given an input query, a recommendation model is trained using user feedback data (e.g., click data) to output a ranked list of items. In real-world systems, besides accuracy, an important consideration for a new model is novelty of its top-k recommendations w.r.t. an existing deployed model. However, novelty of top-k items is a difficult goal to optimize a model for, since it involves a non-differentiable sorting operation on the model's predictions. Moreover, novel items, by definition, do not have any user feedback data. Given the semantic capabilities of large language models, we address these problems using a reinforcement learning (RL) formulation where large language models provide feedback for the novel items. However, given millions of candidate items, the sample complexity of a standard RL algorithm can be prohibitively high. To reduce sample complexity, we reduce the top-k list reward to a set of item-wise rewards and reformulate the state space to consist of tuples such that the action space is reduced to a binary decision; and show that this reformulation results in a significantly lower complexity when the number of items is large. We evaluate the proposed algorithm on improving novelty for a query-ad recommendation task on a large-scale search engine. Compared to supervised finetuning on recent pairs, the proposed RL-based algorithm leads to significant novelty gains with minimal loss in recall. We obtain similar results on the ORCAS query-webpage matching dataset and a product recommendation dataset based on Amazon reviews.

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.

Previous language model pre-training methods have uniformly applied a next-token prediction loss to all training tokens. Challenging this norm, we posit that ''Not all tokens in a corpus are equally important for language model training''. Our initial analysis examines token-level training dynamics of language model, revealing distinct loss patterns for different tokens. Leveraging these insights, we introduce a new language model called Rho-1. Unlike traditional LMs that learn to predict every next token in a corpus, Rho-1 employs Selective Language Modeling (SLM), which selectively trains on useful tokens that aligned with the desired distribution. This approach involves scoring training tokens using a reference model, and then training the language model with a focused loss on tokens with higher scores. When continual continual pretraining on 15B OpenWebMath corpus, Rho-1 yields an absolute improvement in few-shot accuracy of up to 30% in 9 math tasks. After fine-tuning, Rho-1-1B and 7B achieved state-of-the-art results of 40.6% and 51.8% on MATH dataset, respectively - matching DeepSeekMath with only 3% of the pretraining tokens. Furthermore, when continual pretraining on 80B general tokens, Rho-1 achieves 6.8% average enhancement across 15 diverse tasks, increasing both data efficiency and performance of the language model pre-training.

Ziheng Li, Shaohan Huang, Zihan Zhang, Zhi-Hong Deng, Qiang Lou, Haizhen Huang, Jian Jiao, Furu Wei, Weiwei Deng, Qi Zhang. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Named entity recognition (NER) is a crucial task for online advertisement. State-of-the-art solutions leverage pre-trained language models for this task. However, three major challenges remain unresolved: web queries differ from natural language, on which pre-trained models are trained; web queries are short and lack contextual information; and labeled data for NER is scarce. We propose DeepTagger, a knowledge-enhanced NER model for web-based ads queries. The proposed knowledge enhancement framework leverages both model-free and model-based approaches. For model-free enhancement, we collect unlabeled web queries to augment domain knowledge; and we collect web search results to enrich the information of ads queries. We further leverage effective prompting methods to automatically generate labels using large language models such as ChatGPT. Additionally, we adopt a model-based knowledge enhancement method based on adversarial data augmentation. We employ a three-stage training framework to train DeepTagger models.

The dual-encoder has become the de facto architecture for dense retrieval.Typically, it computes the latent representations of the query and document independently, thus failing to fully capture the interactions between the query and document.To alleviate this, recent research has focused on obtaining query-informed document representations.During training, it expands the document with a real query, but during inference, it replaces the real query with a generated one.This inconsistency between training and inference causes the dense retrieval model to prioritize query information while disregarding the document when computing the document representation.Consequently, it performs even worse than the vanilla dense retrieval model because its performance heavily relies on the relevance between the generated queries and the real query.In this paper, we propose a curriculum sampling strategy that utilizes pseudo queries during training and progressively enhances the relevance between the generated query and the real query.By doing so, the retrieval model learns to extend its attention from the document alone to both the document and query, resulting in high-quality queryinformed document representations.Experimental results on both in-domain and out-ofdomain datasets demonstrate that our approach outperforms previous dense retrieval models.

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.

Diffusion models have gained significant attention in the realm of image generation due to their exceptional performance. Their success has been recently expanded to text generation via generating all tokens within a sequence concurrently. However, natural language exhibits a far more pronounced sequential dependency in comparison to images, and the majority of existing language models are trained with a left-to-right auto-regressive approach. To account for the inherent sequential characteristic of natural language, we introduce Auto-Regressive Diffusion (AR-Diffusion). AR-Diffusion ensures that the generation of tokens on the right depends on the generated ones on the left, a mechanism achieved through employing a dynamic number of denoising steps that vary based on token position. This results in tokens on the left undergoing fewer denoising steps than those on the right, thereby enabling them to generate earlier and subsequently influence the generation of tokens on the right. In a series of experiments on various text generation tasks, including text summarization, machine translation, and common sense generation, AR-Diffusion clearly demonstrated its superiority over existing diffusion language models and that it can be $100\times\sim600\times$ faster when achieving comparable results. Our code is available at https://github.com/microsoft/ProphetNet/tree/master/AR-diffusion.

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

Knowledge distillation is an effective way to transfer knowledge from a strong teacher to an efficient student model. Ideally, we expect the better the teacher is, the better the student performs. However, this expectation does not always come true. It is common that a strong teacher model results in a bad student via distillation due to the nonnegligible gap between teacher and student. To bridge the gap, we propose PROD, a PROgressive Distillation method, for dense retrieval. PROD consists of a teacher progressive distillation and a data progressive distillation to gradually improve the student. To alleviate catastrophic forgetting, we introduce a regularization term in each distillation process. We conduct extensive experiments on seven datasets including five widely-used publicly available benchmarks: MS MARCO Passage, TREC Passage 19, TREC Document 19, MS MARCO Document, and Natural Questions, as well as two industry datasets: Bing-Rel and Bing-Ads. PROD achieves the state-of-the-art in the distillation methods for dense retrieval. Our 6-layer student model even surpasses most of the existing 12-layer models on all five public benchmarks. The code and models are released in https://github.com/microsoft/SimXNS.

Xingwei He, Yeyun Gong, A-Long Jin, Weizhen Qi, Hang Zhang, Jian Jiao, Bartuer Zhou, Biao Cheng, Sm Yiu, Nan Duan. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Ting Jiang, Jian Jiao, Shaohan Huang, Zihan Zhang, Deqing Wang, Fuzhen Zhuang, Furu Wei, Haizhen Huang, Denvy Deng, Qi Zhang. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

This paper examines the challenging problem of learning representations of entities and relations in a complex multi-relational knowledge graph.We propose HittER, a Hierarchical Transformer model to jointly learn Entityrelation composition and Relational contextualization based on a source entity's neighborhood.Our proposed model consists of two different Transformer blocks: the bottom block extracts features of each entity-relation pair in the local neighborhood of the source entity and the top block aggregates the relational information from outputs of the bottom block.We further design a masked entity prediction task to balance information from the relational context and the source entity itself.Experimental results show that HittER achieves new stateof-the-art results on multiple link prediction datasets.We additionally propose a simple approach to integrate HittER into BERT and demonstrate its effectiveness on two Freebase factoid question answering datasets.

Deep extreme multi-label learning (XML) requires training deep architectures that can tag a data point with its most relevant subset of labels from an extremely large label set. XML applications such as ad and product recommendation involve labels rarely seen during training but which nevertheless hold the key to recommendations that delight users. Effective utilization of label metadata and high quality predictions for rare labels at the scale of millions of labels are thus key challenges in contemporary XML research. To address these, this paper develops the SiameseXML framework based on a novel probabilistic model that naturally motivates a modular approach melding Siamese architectures with high-capacity extreme classifiers, and a training pipeline that effortlessly scales to tasks with 100 million labels. SiameseXML offers predictions 2–13% more accurate than leading XML methods on public benchmark datasets, as well as in live A/B tests on the Bing search engine, it offers significant gains in click-through-rates, coverage, revenue and other online metrics over state-of-the-art techniques currently in production. Code for SiameseXML is available at https://github.com/Extreme-classification/siamesexml

In this paper, we propose BANG, a new pretraining model to Bridge the gap between Autoregressive (AR) and Non-autoregressive (NAR) Generation. AR and NAR generation can be uniformly regarded as to what extent previous tokens can be attended, and BANG bridges AR and NAR generation through designing a novel model structure for large-scale pre-training. A pretrained BANG model can simultaneously support AR, NAR, and semi-NAR generation to meet different requirements. Experiments on question generation (SQuAD 1.1), summarization (XSum), and dialogue generation (PersonaChat) show that BANG improves NAR and semi-NAR performance significantly as well as attaining comparable performance with strong AR pretrained models. Compared with the semi-NAR strong baselines, BANG achieves absolute improvements of 14.01 and 5.24 in the overall scores of SQuAD 1.1 and XSum, respectively. In addition, BANG achieves absolute improvements of 10.73, 6.39, and 5.90 in the overall scores of SQuAD, XSUM, and PersonaChat compared with the NAR strong baselines, respectively. Our code will be made publicly available.

Zhihao Fan, Yeyun Gong, Dayiheng Liu, Zhongyu Wei, Siyuan Wang, Jian Jiao, Nan Duan, Ruofei Zhang, Xuanjing Huang. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

This paper develops the GalaXC algorithm for Extreme Classification, where the task is to annotate a document with the most relevant subset of labels from an extremely large label set. Extreme classification has been successfully applied to several real world web-scale applications such as web search, product recommendation, query rewriting, etc. GalaXC identifies two critical deficiencies in leading extreme classification algorithms. First, existing approaches generally assume that documents and labels reside in disjoint sets, even though in several applications, labels and documents cohabit the same space. Second, several approaches, albeit scalable, do not utilize various forms of metadata offered by applications, such as label text and label correlations. To remedy these, GalaXC presents a framework that enables collaborative learning over joint document-label graphs at massive scales, in a way that naturally allows various auxiliary sources of information, including label metadata, to be incorporated. GalaXC also introduces a novel label-wise attention mechanism to meld high-capacity extreme classifiers with its framework. An efficient end-to-end implementation of GalaXC is presented that could be trained on a dataset with 50M labels and 97M training documents in less than 100 hours on 4 × V100 GPUs. This allowed GalaXC to not only scale to applications with several millions of labels, but also be up to 18% more accurate than leading deep extreme classifiers, while being upto 2-50 × faster to train and 10 × faster to predict on benchmark datasets. GalaXC is particularly well-suited to warm-start scenarios where predictions need to be made on data points with partially revealed label sets, and was found to be up to 25% more accurate than extreme classification algorithms specifically designed for warm start settings. In A/B tests conducted on the Bing search engine, GalaXC could improve the Click Yield (CY) and coverage by 1.52% and 1.11% respectively. Code for GalaXC is available at https://github.com/Extreme-classification/GalaXC

Pre-trained language models have achieved great success in a wide variety of natural language processing (NLP) tasks, while the superior performance comes with high demand in computational resources, which hinders the application in low-latency information retrieval (IR) systems. To address the problem, we present TwinBERT model, which has two improvements: 1) represent query and document separately using twin-structured encoders and 2) each encoder is a highly compressed BERT-like model with less than one third of the parameters. The former allows document embeddings to be pre-computed offline and cached in memory, which is different from BERT, where the two input sentences are concatenated and encoded together. The change saves large amount of computation time, however, it is still not sufficient for real-time retrieval considering the complexity of BERT model itself. To further reduce computational cost, a compressed multi-layer transformer encoder is proposed with special training strategies as a substitution of the original complex BERT encoder. Lastly, two versions of TwinBERT are developed to combine the query and keyword embeddings for retrieval and relevance tasks correspondingly. Both of them have met the real-time latency requirement and achieve close or on-par performance to BERT-Base model.

Commonsense generation aims at generating plausible everyday scenario description based on a set of provided concepts. Digging the relationship of concepts from scratch is non-trivial, therefore, we retrieve prototypes from external knowledge to assist the understanding of the scenario for better description generation. We integrate two additional modules into the pretrained encoder-decoder model for prototype modeling to enhance the knowledge injection procedure. We conduct experiment on CommonGen benchmark, experimental results show that our method significantly improves the performance on all the metrics.

Rapid advances in GPU hardware and multiple areas of Deep Learning open up a new opportunity for billion-scale information retrieval with exhaustive search. Building on top of the powerful concept of semantic learning, this paper proposes a Recurrent Binary Embedding (RBE) model that learns compact representations for real-time retrieval. The model has the unique ability to refine a base binary vector by progressively adding binary residual vectors to meet the desired accuracy. The refined vector enables efficient implementation of exhaustive similarity computation with bit-wise operations, followed by a near-lossless k-NN selection algorithm, also proposed in this paper. The proposed algorithms are integrated into an end-to-end multi-GPU system that retrieves thousands of top items from over a billion candidates in real-time. The RBE model and the retrieval system were evaluated with data from a major paid search engine. When measured against the state-of-the-art model for binary representation and the full precision model for semantic embedding, RBE significantly outperformed the former, and filled in over 80% of the AUC gap in-between. Experiments comparing with our production retrieval system also demonstrated superior performance. While the primary focus of this paper is to build RBE based on a particular class of semantic models, generalizing to other types is straightforward, as exemplified by two different models at the end of the paper.

Manually crafted combinatorial features have been the "secret sauce" behind many successful models. For web-scale applications, however, the variety and volume of features make these manually crafted features expensive to create, maintain, and deploy. This paper proposes the Deep Crossing model which is a deep neural network that automatically combines features to produce superior models. The input of Deep Crossing is a set of individual features that can be either dense or sparse. The important crossing features are discovered implicitly by the networks, which are comprised of an embedding and stacking layer, as well as a cascade of Residual Units. Deep Crossing is implemented with a modeling tool called the Computational Network Tool Kit (CNTK), powered by a multi-GPU platform. It was able to build, from scratch, two web-scale models for a major paid search engine, and achieve superior results with only a sub-set of the features used in the production models. This demonstrates the potential of using Deep Crossing as a general modeling paradigm to improve existing products, as well as to speed up the development of new models with a fraction of the investment in feature engineering and acquisition of deep domain knowledge.