In this paper, we demonstrate that information retrieval can be accomplished with a single Transformer, in which all information about the corpus is encoded in the parameters of the model. To this end, we introduce the Differentiable Search Index (DSI), a new paradigm that learns a text-to-text model that maps string queries directly to relevant docids; in other words, a DSI model answers queries directly using only its parameters, dramatically simplifying the whole retrieval process. We study variations in how documents and their identifiers are represented, variations in training procedures, and the interplay between models and corpus sizes. Experiments demonstrate that given appropriate design choices, DSI significantly outperforms strong baselines such as dual encoder models. Moreover, DSI demonstrates strong generalization capabilities, outperforming a BM25 baseline in a zero-shot setup.
We argue that current IR metrics, modeled on optimizing user experience, measure too narrow a portion of the IR space. If IR systems are weak, these metrics undersample or completely filter out the deeper documents that need improvement. If IR systems are relatively strong, these metrics undersample deeper relevant documents that could underpin even stronger IR systems, ones that could present content from tens or hundreds of relevant documents in a user-digestible hierarchy or text summary. We reanalyze over 70 TREC tracks from the past 28 years, showing that roughly half undersample top ranked documents and nearly all undersample tail documents. We show that in the 2020 Deep Learning tracks, neural systems were actually near-optimal at top-ranked documents, compared to only modest gains over BM25 on tail documents. Our analysis is based on a simple new systems-oriented metric, 'atomized search length', which is capable of accurately and evenly measuring all relevant documents at any depth.
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.
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.
The world of empirical machine learning (ML) strongly relies on benchmarks in order to determine the relative effectiveness of different algorithms and methods. This paper proposes the notion of "a benchmark lottery" that describes the overall fragility of the ML benchmarking process. The benchmark lottery postulates that many factors, other than fundamental algorithmic superiority, may lead to a method being perceived as superior. On multiple benchmark setups that are prevalent in the ML community, we show that the relative performance of algorithms may be altered significantly simply by choosing different benchmark tasks, highlighting the fragility of the current paradigms and potential fallacious interpretation derived from benchmarking ML methods. Given that every benchmark makes a statement about what it perceives to be important, we argue that this might lead to biased progress in the community. We discuss the implications of the observed phenomena and provide recommendations on mitigating them using multiple machine learning domains and communities as use cases, including natural language processing, computer vision, information retrieval, recommender systems, and reinforcement learning.
State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a block scoring network. We additionally introduce Charformer, a deep Transformer model that integrates GBST and operates on the byte level. Via extensive experiments on English GLUE, multilingual, and noisy text datasets, we show that Charformer outperforms a series of competitive byte-level baselines while generally performing on par and sometimes outperforming subword-based models. Additionally, Charformer is fast, improving the speed of both vanilla byte-level and subword-level Transformers by 28%-100% while maintaining competitive quality. We believe this work paves the way for highly performant token-free models that are trained completely end-to-end.
Self-supervised contrastive representation learning has proved incredibly successful in the vision and natural language domains, enabling state-of-the-art performance with orders of magnitude less labeled data. However, such methods are domain-specific and little has been done to leverage this technique on real-world tabular datasets. We propose SCARF, a simple, widely-applicable technique for contrastive learning, where views are formed by corrupting a random subset of features. When applied to pre-train deep neural networks on the 69 real-world, tabular classification datasets from the OpenML-CC18 benchmark, SCARF not only improves classification accuracy in the fully-supervised setting but does so also in the presence of label noise and in the semi-supervised setting where only a fraction of the available training data is labeled. We show that SCARF complements existing strategies and outperforms alternatives like autoencoders. We conduct comprehensive ablations, detailing the importance of a range of factors.
In the pursuit of a deeper understanding of a model's behaviour, there is recent impetus for developing suites of probes aimed at diagnosing models beyond simple metrics like accuracy or BLEU. This paper takes a step back and asks an important and timely question: how reliable are these diagnostics in providing insight into models and training setups? We critically examine three recent diagnostic tests for pre-trained language models, and find that likelihood-based and representation-based model diagnostics are not yet as reliable as previously assumed. Based on our empirical findings, we also formulate recommendations for practitioners and researchers.
In the era of pre-trained language models, Transformers are the de facto choice of model architectures. While recent research has shown promise in entirely convolutional, or CNN, architectures, they have not been explored using the pre-train-fine-tune paradigm. In the context of language models, are convolutional models competitive to Transformers when pre-trained? This paper investigates this research question and presents several interesting findings. Across an extensive set of experiments on 8 datasets/tasks, we find that CNN-based pre-trained models are competitive and outperform their Transformer counterpart in certain scenarios, albeit with caveats. Overall, the findings outlined in this paper suggest that conflating pre-training and architectural advances is misguided and that both advances should be considered independently. We believe our research paves the way for a healthy amount of optimism in alternative architectures.