Retrieval Augmentation and In-Context Learning

Table of Contents

Sufficient Context: A New Lens on Retrieval Augmented Generation Systems (7 Dec 2024)

Google

we explore ways to reduce hallucinations in RAG systems, including a new selective generation method that leverages sufficient context information for guided abstention.

To divide the dataset, we use an LLM-based autorater to classify instances as sufficient or not. Here, an autorater is a model that evaluates instances based on a property, e.g., a sufficient context autorater.

First, we find models generate incorrect answers on a non-trivial fraction of instances that have sufficient context to answer the query. In other words, open-book QA cannot be solved by improving retrieval alone. Second, when given instances without sufficient context, models tend to hallucinate more than they abstain, especially for multi-hop questions. This finding complements prior work, which shows that LLMs are not robust to noisy retrieval (Yoran et al., 2024; Wu et al., 2024). Third, models generate correct answers in many cases, even when the provided context is insufficient. Surprisingly, this remains true after we filter out questions that the model answers correctly in a closed book (w/o RAG) setting.

We also find that fine-tuning open-source models with sufficient context information does not easily reduce the hallucination rate. Instead, for LLama 3.1 and Mistral 3, fine-tuning can lead to a higher abstention rate at the cost of fewer correct answers.

Inference Scaling for Long-Context Retrieval Augmented Generation (2024-10-05)

Google DeepMind

we investigate inference scaling for retrieval augmented generation (RAG), exploring strategies beyond simply increasing the quantity of knowledge. We focus on two inference scaling strategies: in-context learning and iterative prompting.

increasing inference computation leads to nearly linear gains in RAG performance when optimally allocated, a relationship we describe as the inference scaling laws for RAG.

Inspired by iterative methods (Trivedi et al., 2023; Yoran et al., 2024), we develop iterative demonstration-based RAG (IterDRAG). IterDRAG learns to decompose input queries into simpler sub-queries and answer them using interleaved retrieval. By iteratively retrieving and generating upon sub-queries, LLMs construct reasoning chains that bridge the compositionality gap for multi-hop queries.

we measure computation by considering the total number of input tokens across all iterations, referred to as the effective context length.

We reverse the order of the retrieved documents, placing higher-ranked documents closer to the query (Liu et al., 2024b)

Memory-Augmented LLM Personalization with Short- and Long-Term Memory Coordination (2023)

TBC

ReLLa: Retrieval-enhanced Large Language Models for Lifelong Sequential Behavior Comprehension in Recommendation (2023)

TBC

RUEL: Retrieval-Augmented User Representation with Edge Browser Logs for Sequential Recommendation (2023)

we propose RUEL, a novel retrieval-based sequential recommender that can effectively incorporate external anonymous user behavior data. … We then design a contrastive learning framework with a momentum encoder and a memory bank to retrieve the most relevant and diverse browsing sequences from the full browsing log based on the semantic similarity between user representations. After retrieval, we apply an item-level attentive selector to filter out noisy items and generate refined sequence embeddings for the final predictor.

TBC

Atlas: Few-shot Learning with Retrieval Augmented Language Models (2022)

Atlas retrieves relevant documents based on the current context by using a general-purpose dense retriever using a dual-encoder architecture, based on the Contriever (Izacard et al., 2022). The retrieved documents are processed, along with the current context, by a sequence-to-sequence model using the Fusion-in-Decoder architecture (Izacard & Grave, 2020) that generates the corresponding output. We study the impact of different techniques to train Atlas on its few-shot performance on a range of downstream tasks, including question answering and fact checking. We find that jointly pre-training the components is crucial for few-shot performance.

Retriever. Our retriever module is based on the Contriever (Izacard et al., 2022), an information retrieval technique based on continuous dense embeddings. The Contriever uses a dual-encoder architecture, where the query and documents are embedded independently by a transformer encoder (Huang et al., 2013; Karpukhin et al., 2020). Average pooling is applied over the outputs of the last layer to obtain one vector representation per query or document.

Language model. For the language model, we rely on the T5 sequence-to-sequence architecture (Raffel et al., 2019). We rely on the Fusion-in-Decoder modification of sequence-to-sequence models, and process each document independently in the encoder.

if the language model finds a document useful when generating the output, the retriever objective should encourage the retriever to rank said document higher. This allows us to train models using only query and output pairs from the task of interest, without relying on document annotations.

Attention Distillation (ADist). The first loss that we consider is based on the attention scores of the language model, and is heavily inspired by Izacard & Grave (2021). The main idea is that the cross-attention scores between the input documents and the output, can be used as a proxy of the importance of each input document when generating the output. … average across heads, layers and tokens. … Here, this loss is only used to optimize the parameters of the retriever, and not the language model. When using recent deep learning frameworks, this is achieved by applying a StopGradient operator on pattn.

End-to-end training of Multi-Document Reader and Retriever (EMDR2). Next, we consider the method introduced by Sachan et al. (2021), which is inspired by the expectation-maximization algorithm, treating retrieved documents as latent variables. … Again, only the parameters of the retriever are updated by applying a StopGradient operator

Perplexity Distillation (PDist). Third, …, we want to train the retriever to predict how much each document would improve the language model perplexity of the output, given the query.

Leave-one-out Perplexity Distillation (LOOP). Finally, we propose an objective based on how much worse the prediction of the language model gets, when removing one of the top-k retrieved documents.

Efficient retriever fine-tuning: Re-ranking. A second strategy is to retrieve a larger number of documents L with the retriever, and to re-embed and rerank these documents with the up-to-date retriever, and pass the resulting top-K to the language model.

Query-side fine-tuning. Finally, the last strategy is to decouple the encoding of the queries and documents. In this case, we fix the parameters corresponding to the document encoder, and only train the parameters corresponding to the query encoder. Thus, the embeddings of documents are fixed, and we do not need to refresh the index, and thus there is no computational overhead. As we will see in practice, the impact of fixing the documents encoder varies greatly for different tasks when a large training dataset is available. For most of the few-shot settings that we consider, query-side finetuning does not have large performance impact, and sometimes even slightly improves performance.

RAVEN: In-Context Learning with Retrieval Augmented Encoder-Decoder Language Models (2023)

We find that ATLAS exhibits a certain in-context learning ability; however, due to a mismatch between pretraining and testing and a limited context length—issues that are common to existing encoderdecoder LMs trained with masked language modeling—its few-shot performance is not stable and providing more than, e.g., 8-shot, examples does not lead to further improvement.

While there is growing interest in this area, most studies have focused on incontext learning with decoder-only LMs, e.g., GPT-3 (Brown et al., 2020). However, bidirectional LMs like BERT (Devlin et al., 2019) and T5 (Raffel et al., 2020) have been shown to achieve superior performance on various natural language understanding tasks, indicating that they may offer unique advantages for in-context learning as well. … For instance, Patel et al. (2023) demonstrate that bidirectional models can outperform decoder-only LMs of a similar scale regarding in-context learning; however, there is still a significant performance gap compared to decoder-only models on a much larger scale.

While there has been some research on in-context learning with retrieval-augmented decoder-only LMs, which can be straightforwardly implemented by concatenating retrieved passages with the query as the input of the LM (Mallen et al., 2022; Shi et al., 2023; Khattab et al., 2022), in-context learning with retrieval-augmented encoder-decoder LMs, such as ATLAS, remains unexplored

TBC

RETRIEVAL MEETS LONG CONTEXT LARGE LANGUAGE MODELS (2023)

we find that LLM with 4K context window using simple retrieval-augmentation at generation can achieve comparable performance to finetuned LLM with 16K context window via positional interpolation on long context tasks, while taking much less computation. More importantly, we demonstrate that retrieval can significantly improve the performance of LLMs regardless of their extended context window sizes

TBC

In-Context Retrieval-Augmented Language Models (2023)

incontext RALM: leaving the LM architecture unchanged and prepending grounding documents to the input. We show that in-context RALM which uses off-the-shelf general purpose retrievers provides surprisingly large LM gains across five diverse corpora.

MAKING RETRIEVAL-AUGMENTED LANGUAGE MODELS ROBUST TO IRRELEVANT CONTEXT (2023)

TBC

UNDERSTANDING RETRIEVAL AUGMENTATION FOR LONG-FORM QUESTION ANSWERING (2023)

TBC

RET-LLM: Towards a General Read-Write Memory for Large Language Models (2023)

TBC

Shall We Pretrain Autoregressive Language Models with Retrieval? A Comprehensive Study (2023)

TBC

RecMind: Large Language Model Powered Agent For Recommendation (2023)

TBC

Generative Slate Recommendation with Reinforcement Learning (2023)

TBC

LlamaRec: Two-Stage Recommendation using Large Language Models for Ranking (2023)

TBC

Integrating Summarization and Retrieval for Enhanced Personalization via Large Language Models (2023)

TBC

End-to-End Training of Multi-Document Reader and Retriever for Open-Domain Question Answering (2021)

TBC

Augmented Language Models: a Survey (2023)

TBC

You can’t pick your neighbors, or can you? When and how to rely on retrieval in the kNN-LM (2022)

we explore the importance of lexical and semantic matching in the context of items retrieved by kNNLM. We find two trends: (1) the presence of large overlapping n-grams between the datastore and evaluation set plays an important factor in strong performance, even when the datastore is derived from the training data; and (2) the kNN-LM is most beneficial when retrieved items have high semantic similarity with the query. example TBC

Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? (2022)

TBC

Ground-Truth Labels Matter: A Deeper Look into Input-Label Demonstrations (2022)

TBC

LARGER LANGUAGE MODELS DO IN-CONTEXT LEARNING DIFFERENTLY (2023)

TBC

SAIL: Search-Augmented Instruction Learning (2023)

search engine –> duckduckgo

TBC

RECITATION-AUGMENTED LANGUAGE MODELS (2023)

by Google research

TBC

BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage (2022)

by Meta AI –> Mojeek

TBC

Internet-Augmented Dialogue Generation (2021)

by Facebook AI research –> bing search api

TBC

FRESHLLMS: REFRESHING LARGE LANGUAGE MODELS WITH SEARCH ENGINE AUGMENTATION (2023)

by Google and OpenAI

TBC

Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts for Open-Domain QA? (2023)

we construct datasets with conflicting contexts, where each question is paired with both generated and retrieved contexts, yet only one of them contains the correct answer. Our experiments reveal a significant bias in LLMs (GPT-4/3.5 and Llama2) towards generated contexts, even when they provide incorrect information. We further identify two key factors contributing to this bias: i) contexts generated by LLMs typically show greater similarity to the questions, increasing their likelihood of selection; ii) the segmentation process used in retrieved contexts disrupts their completeness, thereby hindering their full utilization in LLMs.

TBC