PROD: Progressive Distillation for Dense Retrieval

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. However, this expectation does not always come true. It is common that a better 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. We conduct extensive experiments on five widely-used benchmarks, MS MARCO Passage, TREC Passage 19, TREC Document 19, MS MARCO Document and Natural Questions, where PROD achieves the state-of-the-art within the distillation methods for dense retrieval. The code and models will be released.

Joint Generator-Ranker Learning for Natural Language Generation

Due to exposure bias, most existing natural language generation (NLG) models trained by maximizing the likelihood objective predict poor text results during the inference stage. In this paper, to tackle this problem, we revisit the generate-then-rank framework and propose a joint generator-ranker (JGR) training algorithm for text generation tasks. In JGR, the generator model is trained by maximizing two objectives: the likelihood of the training corpus and the expected reward given by the ranker model. Meanwhile, the ranker model takes input samples from the generator model and learns to distinguish good samples from the generation pool. The generator and ranker models are alternately optimized till convergence. In the empirical study, the proposed JGR model achieves new state-of-the-art performance on five public benchmarks covering three popular generation tasks: summarization, question generation, and response generation. We will make code, data, and models available at https://github.com/microsoft/AdvNLG.

A Self-Paced Mixed Distillation Method for Non-Autoregressive Generation

Non-Autoregressive generation is a sequence generation paradigm, which removes the dependency between target tokens. It could efficiently reduce the text generation latency with parallel decoding in place of token-by-token sequential decoding. However, due to the known multi-modality problem, Non-Autoregressive (NAR) models significantly under-perform Auto-regressive (AR) models on various language generation tasks. Among the NAR models, BANG is the first large-scale pre-training model on English un-labeled raw text corpus. It considers different generation paradigms as its pre-training tasks including Auto-regressive (AR), Non-Autoregressive (NAR), and semi-Non-Autoregressive (semi-NAR) information flow with multi-stream strategy. It achieves state-of-the-art performance without any distillation techniques. However, AR distillation has been shown to be a very effective solution for improving NAR performance. In this paper, we propose a novel self-paced mixed distillation method to further improve the generation quality of BANG. Firstly, we propose the mixed distillation strategy based on the AR stream knowledge. Secondly, we encourage the model to focus on the samples with the same modality by self-paced learning. The proposed self-paced mixed distillation algorithm improves the generation quality and has no influence on the inference latency. We carry out extensive experiments on summarization and question generation tasks to validate the effectiveness. To further illustrate the commercial value of our approach, we conduct experiments on three generation tasks in real-world advertisements applications. Experimental results on commercial data show the effectiveness of the proposed model. Compared with BANG, it achieves significant BLEU score improvement. On the other hand, compared with auto-regressive generation method, it achieves more than 7x speedup.

DialogVED: A Pre-trained Latent Variable Encoder-Decoder Model for Dialog Response Generation

Dialog response generation in open domain is an important research topic where the main challenge is to generate relevant and diverse responses. In this paper, we propose a new dialog pre-training framework called DialogVED, which introduces continuous latent variables into the enhanced encoder-decoder pre-training framework to increase the relevance and diversity of responses. With the help of a large dialog corpus (Reddit), we pre-train the model using the following 4 tasks, used in training language models (LMs) and Variational Autoencoders (VAEs) literature: 1) masked language model; 2) response generation; 3) bag-of-words prediction; and 4) KL divergence reduction. We also add additional parameters to model the turn structure in dialogs to improve the performance of the pre-trained model. We conduct experiments on PersonaChat, DailyDialog, and DSTC7-AVSD benchmarks for response generation. Experimental results show that our model achieves the new state-of-the-art results on all these datasets.

Distill-VQ: Learning Retrieval Oriented Vector Quantization By Distilling Knowledge from Dense Embeddings

Vector quantization (VQ) based ANN indexes, such as Inverted File System (IVF) and Product Quantization (PQ), have been widely applied to embedding based document retrieval thanks to the competitive time and memory efficiency. Originally, VQ is learned to minimize the reconstruction loss, i.e., the distortions between the original dense embeddings and the reconstructed embeddings after quantization. Unfortunately, such an objective is inconsistent with the goal of selecting ground-truth documents for the input query, which may cause severe loss of retrieval quality. Recent works identify such a defect, and propose to minimize the retrieval loss through contrastive learning. However, these methods intensively rely on queries with ground-truth documents, whose performance is limited by the insufficiency of labeled data. In this paper, we propose Distill-VQ, which unifies the learning of IVF and PQ within a knowledge distillation framework. In Distill-VQ, the dense embeddings are leveraged as "teachers", which predict the query's relevance to the sampled documents. The VQ modules are treated as the "students", which are learned to reproduce the predicted relevance, such that the reconstructed embeddings may fully preserve the retrieval result of the dense embeddings. By doing so, Distill-VQ is able to derive substantial training signals from the massive unlabeled data, which significantly contributes to the retrieval quality. We perform comprehensive explorations for the optimal conduct of knowledge distillation, which may provide useful insights for the learning of VQ based ANN index. We also experimentally show that the labeled data is no longer a necessity for high-quality vector quantization, which indicates Distill-VQ's strong applicability in practice.

CodeRetriever: Unimodal and Bimodal Contrastive Learning

In this paper, we propose the CodeRetriever model, which combines the unimodal and bimodal contrastive learning to train function-level code semantic representations, specifically for the code search task. For unimodal contrastive learning, we design a semantic-guided method to build positive code pairs based on the documentation and function name. For bimodal contrastive learning, we leverage the documentation and in-line comments of code to build text-code pairs. Both contrastive objectives can fully leverage the large-scale code corpus for pre-training. Experimental results on several public benchmarks, (i.e., CodeSearch, CoSQA, etc.) demonstrate the effectiveness of CodeRetriever in the zero-shot setting. By fine-tuning with domain/language specified downstream data, CodeRetriever achieves the new state-of-the-art performance with significant improvement over existing code pre-trained models. We will make the code, model checkpoint, and constructed datasets publicly available.

Adversarial Retriever-Ranker for dense text retrieval

Current dense text retrieval models face two typical challenges. First, it adopts a siamese dual-encoder architecture to encode query and document independently for fast indexing and searching, whereas neglecting the finer-grained term-wise interactions. This results in a sub-optimal recall performance. Second, it highly relies on a negative sampling technique to build up the negative documents in its contrastive loss. To address these challenges, we present Adversarial Retriever-Ranker (AR2), which consists of a dual-encoder retriever plus a cross-encoder ranker. The two models are jointly optimized according to a minimax adversarial objective: the retriever learns to retrieve negative documents to cheat the ranker, while the ranker learns to rank a collection of candidates including both the ground-truth and the retrieved ones, as well as providing progressive direct feedback to the dual-encoder retriever. Through this adversarial game, the retriever gradually produces harder negative documents to train a better ranker, whereas the cross-encoder ranker provides progressive feedback to improve retriever. We evaluate AR2 on three benchmarks. Experimental results show that AR2 consistently and significantly outperforms existing dense retriever methods and achieves new state-of-the-art results on all of them. This includes the improvements on Natural Questions R@5 to 77.9%(+2.1%), TriviaQA R@5 to 78.2%(+1.4), and MS-MARCO MRR@10 to 39.5%(+1.3%). We will make our code, models, and data publicly available.

Contextual Fine-to-Coarse Distillation for Coarse-grained Response Selection in Open-Domain Conversations

We study the problem of coarse-grained response selection in retrieval-based dialogue systems. The problem is equally important with fine-grained response selection, but is less explored in existing literature. In this paper, we propose a Contextual Fine-to-Coarse (CFC) distilled model for coarse-grained response selection in open-domain conversations. In our CFC model, dense representations of query, candidate response and corresponding context is learned based on the multi-tower architecture, and more expressive knowledge learned from the one-tower architecture (fine-grained) is distilled into the multi-tower architecture (coarse-grained) to enhance the performance of the retriever. To evaluate the performance of our proposed model, we construct two new datasets based on the Reddit comments dump and Twitter corpus. Extensive experimental results on the two datasets show that the proposed methods achieve a significant improvement over all evaluation metrics compared with traditional baseline methods.

KFCNet: Knowledge Filtering and Contrastive Learning Network for Generative Commonsense Reasoning

Pre-trained language models have led to substantial gains over a broad range of natural language processing (NLP) tasks, but have been shown to have limitations for natural language generation tasks with high-quality requirements on the output, such as commonsense generation and ad keyword generation. In this work, we present a novel Knowledge Filtering and Contrastive learning Network (KFCNet) which references external knowledge and achieves better generation performance. Specifically, we propose a BERT-based filter model to remove low-quality candidates, and apply contrastive learning separately to each of the encoder and decoder, within a general encoder--decoder architecture. The encoder contrastive module helps to capture global target semantics during encoding, and the decoder contrastive module enhances the utility of retrieved prototypes while learning general features. Extensive experiments on the CommonGen benchmark show that our model outperforms the previous state of the art by a large margin: +6.6 points (42.5 vs. 35.9) for BLEU-4, +3.7 points (33.3 vs. 29.6) for SPICE, and +1.3 points (18.3 vs. 17.0) for CIDEr. We further verify the effectiveness of the proposed contrastive module on ad keyword generation, and show that our model has potential commercial value.