Jinfeng Rao

Sanket Vaibhav Mehta, Jai Gupta, Yi Tay et al. · deepmind

Differentiable Search Indices (DSIs) encode a corpus of documents in model parameters and use the same model to answer user queries directly. Despite the strong performance of DSI models, deploying them in situations where the corpus changes over time is computationally expensive because reindexing the corpus requires re-training the model. In this work, we introduce DSI++, a continual learning challenge for DSI to incrementally index new documents while being able to answer queries related to both previously and newly indexed documents. Across different model scales and document identifier representations, we show that continual indexing of new documents leads to considerable forgetting of previously indexed documents. We also hypothesize and verify that the model experiences forgetting events during training, leading to unstable learning. To mitigate these issues, we investigate two approaches. The first focuses on modifying the training dynamics. Flatter minima implicitly alleviate forgetting, so we optimize for flatter loss basins and show that the model stably memorizes more documents ($+12\%$). Next, we introduce a generative memory to sample pseudo-queries for documents and supplement them during continual indexing to prevent forgetting for the retrieval task. Extensive experiments on novel continual indexing benchmarks based on Natural Questions (NQ) and MS MARCO demonstrate that our proposed solution mitigates forgetting significantly. Concretely, it improves the average Hits@10 by $+21.1\%$ over competitive baselines for NQ and requires $6$ times fewer model updates compared to re-training the DSI model for incrementally indexing five corpora in a sequence.

Yi Tay, Jason Wei, Hyung Won Chung et al.

Scaling language models improves performance but comes with significant computational costs. This paper proposes UL2R, a method that substantially improves existing language models and their scaling curves with a relatively tiny amount of extra compute. The key idea is to continue training a state-of-the-art large language model (e.g., PaLM) on a few more steps with UL2's mixture-of-denoiser objective. We show that, with almost negligible extra computational costs and no new sources of data, we are able to substantially improve the scaling properties of large language models on downstream metrics. In this paper, we continue training PaLM with UL2R, introducing a new set of models at 8B, 62B, and 540B scale which we call U-PaLM. Impressively, at 540B scale, we show an approximately 2x computational savings rate where U-PaLM achieves the same performance as the final PaLM 540B model at around half its computational budget (i.e., saving $\sim$4.4 million TPUv4 hours). We further show that this improved scaling curve leads to 'emergent abilities' on challenging BIG-Bench tasks -- for instance, U-PaLM does much better than PaLM on some tasks or demonstrates better quality at much smaller scale (62B as opposed to 540B). Overall, we show that U-PaLM outperforms PaLM on many few-shot setups, i.e., English NLP tasks (e.g., commonsense reasoning, question answering), reasoning tasks with chain-of-thought (e.g., GSM8K), multilingual tasks (MGSM, TydiQA), MMLU and challenging BIG-Bench tasks. Finally, we provide qualitative examples showing the new capabilities of U-PaLM for single and multi-span infilling.

Yi Tay, Mostafa Dehghani, Samira Abnar et al.

There have been a lot of interest in the scaling properties of Transformer models. However, not much has been done on the front of investigating the effect of scaling properties of different inductive biases and model architectures. Do model architectures scale differently? If so, how does inductive bias affect scaling behaviour? How does this influence upstream (pretraining) and downstream (transfer)? This paper conducts a systematic study of scaling behaviour of ten diverse model architectures such as Transformers, Switch Transformers, Universal Transformers, Dynamic convolutions, Performers, and recently proposed MLP-Mixers. Via extensive experiments, we show that (1) architecture is an indeed an important consideration when performing scaling and (2) the best performing model can fluctuate at different scales. We believe that the findings outlined in this work has significant implications to how model architectures are currently evaluated in the community.

Josh Beal, Eric Kim, Jinfeng Rao et al.

While multi-modal Visual Language Models (VLMs) have demonstrated significant success across various domains, the integration of VLMs into recommendation and retrieval systems remains a challenge, due to issues like training objective discrepancies and serving efficiency bottlenecks. This paper introduces PinCLIP, a large-scale visual representation learning approach developed to enhance retrieval and ranking models at Pinterest by leveraging VLMs to learn image-text alignment. We propose a novel hybrid Vision Transformer architecture that utilizes a VLM backbone and a hybrid fusion mechanism to capture multi-modality content representation at varying granularities. Beyond standard image-to-text alignment objectives, we introduce a neighbor alignment objective to model the cross-fusion of multi-modal representations within the Pinterest Pin-Board graph. Offline evaluations show that PinCLIP outperforms state-of-the-art baselines, such as Qwen, by 20% in multi-modal retrieval tasks. Online A/B testing demonstrates significant business impact, including substantial engagement gains across all major surfaces in Pinterest. Notably, PinCLIP significantly addresses the "cold-start" problem, enhancing fresh content distribution with a 15% Repin increase in organic content and 8.7% higher click for new Ads.

Yi Tay, Mostafa Dehghani, Samira Abnar et al.

Transformers do not scale very well to long sequence lengths largely because of quadratic self-attention complexity. In the recent months, a wide spectrum of efficient, fast Transformers have been proposed to tackle this problem, more often than not claiming superior or comparable model quality to vanilla Transformer models. To this date, there is no well-established consensus on how to evaluate this class of models. Moreover, inconsistent benchmarking on a wide spectrum of tasks and datasets makes it difficult to assess relative model quality amongst many models. This paper proposes a systematic and unified benchmark, LRA, specifically focused on evaluating model quality under long-context scenarios. Our benchmark is a suite of tasks consisting of sequences ranging from $1K$ to $16K$ tokens, encompassing a wide range of data types and modalities such as text, natural, synthetic images, and mathematical expressions requiring similarity, structural, and visual-spatial reasoning. We systematically evaluate ten well-established long-range Transformer models (Reformers, Linformers, Linear Transformers, Sinkhorn Transformers, Performers, Synthesizers, Sparse Transformers, and Longformers) on our newly proposed benchmark suite. LRA paves the way towards better understanding this class of efficient Transformer models, facilitates more research in this direction, and presents new challenging tasks to tackle. Our benchmark code will be released at https://github.com/google-research/long-range-arena.

Faye Zhang, Qianyu Cheng, Jasmine Wan et al.

Large Language Models are fundamentally reshaping content discovery through AI-native search systems such as ChatGPT, Gemini, and Claude. Unlike traditional search engines that match keywords to documents, these systems infer user intent, synthesize multimodal evidence, and generate contextual answers directly on the search page, introducing a paradigm shift from Search Engine Optimization (SEO) to Generative Engine Optimization (GEO). For visual content platforms hosting billions of assets, this poses an acute challenge: individual images lack the semantic depth and authority signals that generative search prioritizes, risking disintermediation as user needs are satisfied in-place without site visits. We present Pinterest GEO, a production-scale framework that pioneers reverse search design: rather than generating generic image captions describing what content is, we fine-tune Vision-Language Models (VLMs) to predict what users would actually search for, augmented this with AI agents that mine real-time internet trends to capture emerging search demand. These VLM-generated queries then drive construction of semantically coherent Collection Pages via multimodal embeddings, creating indexable aggregations optimized for generative retrieval. Finally, we employ hybrid VLM and two-tower ANN architectures to build authority-aware interlinking structures that propagate signals across billions of visual assets. Deployed at scale across billions of images and tens of millions of collections, GEO delivers 20\% organic traffic growth contributing to multi-million monthly active user (MAU) growth, demonstrating a principled pathway for visual platforms to thrive in the generative search era.

Han Wang, Mukuntha Narayanan Sundararaman, Onur Gungor et al.

To improve relevance scoring on Pinterest Search, we integrate Large Language Models (LLMs) into our search relevance model, leveraging carefully designed text representations to predict the relevance of Pins effectively. Our approach uses search queries alongside content representations that include captions extracted from a generative visual language model. These are further enriched with link-based text data, historically high-quality engaged queries, user-curated boards, Pin titles and Pin descriptions, creating robust models for predicting search relevance. We use a semi-supervised learning approach to efficiently scale up the amount of training data, expanding beyond the expensive human labeled data available. By utilizing multilingual LLMs, our system extends training data to include unseen languages and domains, despite initial data and annotator expertise being confined to English. Furthermore, we distill from the LLM-based model into real-time servable model architectures and features. We provide comprehensive offline experimental validation for our proposed techniques and demonstrate the gains achieved through the final deployed system at scale.

Faye Zhang, Jasmine Wan, Qianyu Cheng et al.

Online platforms like Pinterest hosting vast content collections traditionally rely on manual curation or user-generated search logs to create keyword landing pages (KLPs) -- topic-centered collection pages that serve as entry points for content discovery. While manual curation ensures quality, it doesn't scale to millions of collections, and search log approaches result in limited topic coverage and imprecise content matching. In this paper, we present PinLanding, a novel content-first architecture that transforms the way platforms create topical collections. Instead of deriving topics from user behavior, our system employs a multi-stage pipeline combining vision-language model (VLM) for attribute extraction, large language model (LLM) for topic generation, and a CLIP-based dual-encoder architecture for precise content matching. Our model achieves 99.7% Recall@10 on Fashion200K benchmark, demonstrating strong attribute understanding capabilities. In production deployment for search engine optimization with 4.2 million shopping landing pages, the system achieves a 4X increase in topic coverage and 14.29% improvement in collection attribute precision over the traditional search log-based approach via human evaluation. The architecture can be generalized beyond search traffic to power various user experiences, including content discovery and recommendations, providing a scalable solution to transform unstructured content into curated topical collections across any content domain.

Vamsi Aribandi, Yi Tay, Tal Schuster et al.

Despite the recent success of multi-task learning and transfer learning for natural language processing (NLP), few works have systematically studied the effect of scaling up the number of tasks during pre-training. Towards this goal, this paper introduces ExMix (Extreme Mixture): a massive collection of 107 supervised NLP tasks across diverse domains and task-families. Using ExMix, we study the effect of multi-task pre-training at the largest scale to date, and analyze co-training transfer amongst common families of tasks. Through this analysis, we show that manually curating an ideal set of tasks for multi-task pre-training is not straightforward, and that multi-task scaling can vastly improve models on its own. Finally, we propose ExT5: a model pre-trained using a multi-task objective of self-supervised span denoising and supervised ExMix. Via extensive experiments, we show that ExT5 outperforms strong T5 baselines on SuperGLUE, GEM, Rainbow, Closed-Book QA tasks, and several tasks outside of ExMix. ExT5 also significantly improves sample efficiency while pre-training.

Sanket Vaibhav Mehta, Jinfeng Rao, Yi Tay et al.

Data-to-text generation focuses on generating fluent natural language responses from structured meaning representations (MRs). Such representations are compositional and it is costly to collect responses for all possible combinations of atomic meaning schemata, thereby necessitating few-shot generalization to novel MRs. In this work, we systematically study the compositional generalization of the state-of-the-art T5 models in few-shot data-to-text tasks. We show that T5 models fail to generalize to unseen MRs, and we propose a template-based input representation that considerably improves the model's generalization capability. To further improve the model's performance, we propose an approach based on self-training using fine-tuned BLEURT for pseudo response selection. On the commonly-used SGD and Weather benchmarks, the proposed self-training approach improves tree accuracy by 46%+ and reduces the slot error rates by 73%+ over the strong T5 baselines in few-shot settings.

Yi Tay, Mostafa Dehghani, Jinfeng Rao et al.

There remain many open questions pertaining to the scaling behaviour of Transformer architectures. These scaling decisions and findings can be critical, as training runs often come with an associated computational cost which have both financial and/or environmental impact. The goal of this paper is to present scaling insights from pretraining and finetuning Transformers. While Kaplan et al. presents a comprehensive study of the scaling behaviour of Transformer language models, the scope is only on the upstream (pretraining) loss. Therefore, it is still unclear if these set of findings transfer to downstream task within the context of the pretrain-finetune paradigm. The key findings of this paper are as follows: (1) we show that aside from only the model size, model shape matters for downstream fine-tuning, (2) scaling protocols operate differently at different compute regions, (3) widely adopted T5-base and T5-large sizes are Pareto-inefficient. To this end, we present improved scaling protocols whereby our redesigned models achieve similar downstream fine-tuning quality while having 50\% fewer parameters and training 40\% faster compared to the widely adopted T5-base model. We publicly release over 100 pretrained checkpoints of different T5 configurations to facilitate future research and analysis.

Anusha Balakrishnan, Jinfeng Rao, Kartikeya Upasani et al.

Generating fluent natural language responses from structured semantic representations is a critical step in task-oriented conversational systems. Avenues like the E2E NLG Challenge have encouraged the development of neural approaches, particularly sequence-to-sequence (Seq2Seq) models for this problem. The semantic representations used, however, are often underspecified, which places a higher burden on the generation model for sentence planning, and also limits the extent to which generated responses can be controlled in a live system. In this paper, we (1) propose using tree-structured semantic representations, like those used in traditional rule-based NLG systems, for better discourse-level structuring and sentence-level planning; (2) introduce a challenging dataset using this representation for the weather domain; (3) introduce a constrained decoding approach for Seq2Seq models that leverages this representation to improve semantic correctness; and (4) demonstrate promising results on our dataset and the E2E dataset.

Yi Tay, Aston Zhang, Luu Anh Tuan et al.

Many state-of-the-art neural models for NLP are heavily parameterized and thus memory inefficient. This paper proposes a series of lightweight and memory efficient neural architectures for a potpourri of natural language processing (NLP) tasks. To this end, our models exploit computation using Quaternion algebra and hypercomplex spaces, enabling not only expressive inter-component interactions but also significantly ($75\%$) reduced parameter size due to lesser degrees of freedom in the Hamilton product. We propose Quaternion variants of models, giving rise to new architectures such as the Quaternion attention Model and Quaternion Transformer. Extensive experiments on a battery of NLP tasks demonstrates the utility of proposed Quaternion-inspired models, enabling up to $75\%$ reduction in parameter size without significant loss in performance.

Yi Tay, Shuohang Wang, Luu Anh Tuan et al.

This paper tackles the problem of reading comprehension over long narratives where documents easily span over thousands of tokens. We propose a curriculum learning (CL) based Pointer-Generator framework for reading/sampling over large documents, enabling diverse training of the neural model based on the notion of alternating contextual difficulty. This can be interpreted as a form of domain randomization and/or generative pretraining during training. To this end, the usage of the Pointer-Generator softens the requirement of having the answer within the context, enabling us to construct diverse training samples for learning. Additionally, we propose a new Introspective Alignment Layer (IAL), which reasons over decomposed alignments using block-based self-attention. We evaluate our proposed method on the NarrativeQA reading comprehension benchmark, achieving state-of-the-art performance, improving existing baselines by $51\%$ relative improvement on BLEU-4 and $17\%$ relative improvement on Rouge-L. Extensive ablations confirm the effectiveness of our proposed IAL and CL components.

Peng Shi, Jinfeng Rao, Jimmy Lin

This paper explores the problem of ranking short social media posts with respect to user queries using neural networks. Instead of starting with a complex architecture, we proceed from the bottom up and examine the effectiveness of a simple, word-level Siamese architecture augmented with attention-based mechanisms for capturing semantic "soft" matches between query and post tokens. Extensive experiments on datasets from the TREC Microblog Tracks show that our simple models not only achieve better effectiveness than existing approaches that are far more complex or exploit a more diverse set of relevance signals, but are also much faster. Implementations of our samCNN (Simple Attention-based Matching CNN) models are shared with the community to support future work.

Jinfeng Rao, Wei Yang, Yuhao Zhang et al.

Despite substantial interest in applications of neural networks to information retrieval, neural ranking models have only been applied to standard ad hoc retrieval tasks over web pages and newswire documents. This paper proposes MP-HCNN (Multi-Perspective Hierarchical Convolutional Neural Network) a novel neural ranking model specifically designed for ranking short social media posts. We identify document length, informal language, and heterogeneous relevance signals as features that distinguish documents in our domain, and present a model specifically designed with these characteristics in mind. Our model uses hierarchical convolutional layers to learn latent semantic soft-match relevance signals at the character, word, and phrase levels. A pooling-based similarity measurement layer integrates evidence from multiple types of matches between the query, the social media post, as well as URLs contained in the post. Extensive experiments using Twitter data from the TREC Microblog Tracks 2011--2014 show that our model significantly outperforms prior feature-based as well and existing neural ranking models. To our best knowledge, this paper presents the first substantial work tackling search over social media posts using neural ranking models.

Royal Sequiera, Gaurav Baruah, Zhucheng Tu et al.

Most work on natural language question answering today focuses on answer selection: given a candidate list of sentences, determine which contains the answer. Although important, answer selection is only one stage in a standard end-to-end question answering pipeline. This paper explores the effectiveness of convolutional neural networks (CNNs) for answer selection in an end-to-end context using the standard TrecQA dataset. We observe that a simple idf-weighted word overlap algorithm forms a very strong baseline, and that despite substantial efforts by the community in applying deep learning to tackle answer selection, the gains are modest at best on this dataset. Furthermore, it is unclear if a CNN is more effective than the baseline in an end-to-end context based on standard retrieval metrics. To further explore this finding, we conducted a manual user evaluation, which confirms that answers from the CNN are detectably better than those from idf-weighted word overlap. This result suggests that users are sensitive to relatively small differences in answer selection quality.

Jinfeng Rao, Hua He, Haotian Zhang et al.

Time is an important relevance signal when searching streams of social media posts. The distribution of document timestamps from the results of an initial query can be leveraged to infer the distribution of relevant documents, which can then be used to rerank the initial results. Previous experiments have shown that kernel density estimation is a simple yet effective implementation of this idea. This paper explores an alternative approach to mining temporal signals with recurrent neural networks. Our intuition is that neural networks provide a more expressive framework to capture the temporal coherence of neighboring documents in time. To our knowledge, we are the first to integrate lexical and temporal signals in an end-to-end neural network architecture, in which existing neural ranking models are used to generate query-document similarity vectors that feed into a bidirectional LSTM layer for temporal modeling. Our results are mixed: existing neural models for document ranking alone yield limited improvements over simple baselines, but the integration of lexical and temporal signals yield significant improvements over competitive temporal baselines.

Jinfeng Rao, Ferhan Ture, Hua He et al.

We tackle the novel problem of navigational voice queries posed against an entertainment system, where viewers interact with a voice-enabled remote controller to specify the program to watch. This is a difficult problem for several reasons: such queries are short, even shorter than comparable voice queries in other domains, which offers fewer opportunities for deciphering user intent. Furthermore, ambiguity is exacerbated by underlying speech recognition errors. We address these challenges by integrating word- and character-level representations of the queries and by modeling voice search sessions to capture the contextual dependencies in query sequences. Both are accomplished with a probabilistic framework in which recurrent and feedforward neural network modules are organized in a hierarchical manner. From a raw dataset of 32M voice queries from 2.5M viewers on the Comcast Xfinity X1 entertainment system, we extracted data to train and test our models. We demonstrate the benefits of our hybrid representation and context-aware model, which significantly outperforms models without context as well as the current deployed product.