More Like this

1. Propagating Knowledge Updates to LMs Through Distillation

   Padmanabhan, Shankar; Onoe, Yasumasa; Zhang, Michael JQ; Durrett, Greg; Choi, Eunsol (December 2023, Advances in neural information processing systems)

   Modern language models have the capacity to store and use immense amounts of knowledge about real-world entities, but it remains unclear how to update such knowledge stored in model parameters. While prior methods for updating knowledge in LMs successfully inject atomic facts, updated LMs fail to make inferences based on injected facts. In this work, we demonstrate that a context distillation-based approach can both impart knowledge about entities and propagate that knowledge to enable broader inferences. Our approach consists of two stages: transfer set generation and distillation on the transfer set. We first generate a transfer set by prompting a language model to generate continuations from the entity definition. Then, we update the model parameters so that the distribution of the LM (the student) matches the distribution of the LM conditioned on the definition (the teacher) on the transfer set. Our experiments demonstrate that this approach is more effective at propagating knowledge updates than fine-tuning and other gradient-based knowledge-editing methods. Moreover, it does not compromise performance in other contexts, even when injecting the definitions of up to 150 entities at once. 

   more » « less
2. Inspecting and Editing Knowledge Representations in Language Models

   Hernandez, Evan; Li, Belinda; Andreas, Jacob (October 2024, Conference on Language Models)

   Neural language models (LMs) represent facts about the world described by text. Sometimes these facts derive from training data (in most LMs, a representation of the word banana encodes the fact that bananas are fruits). Sometimes facts derive from input text itself (a representation of the sentence I poured out the bottle encodes the fact that the bottle became empty). We describe REMEDI, a method for learning to map statements in natural language to fact encodings in an LM’s internal representation system. REMEDI encodings can be used as knowledge editors: when added to LM hidden representations, they modify downstream generation to be consistent with new facts. REMEDI encodings may also be used as probes: when compared to LM representations, they reveal which properties LMs already attribute to mentioned entities, in some cases making it possible to predict when LMs will generate outputs that conflict with background knowledge or input text. REMEDI thus links work on probing, prompting, and LM editing, and offers steps toward general tools for fine-grained inspection and control of knowledge in LMs. 

   more » « less
3. QA-GNN: Reasoning with Language Models and Knowledge Graphs for Question Answering

   Yasunaga, Michihiro; Ren, Hongyu; Bosselut, Antoine; Liang, Percy; Leskovec, Jure (January 2021, North American Chapter of the Association for Computational Linguistics (NAACL))

   null (Ed.)

   The problem of answering questions using knowledge from pre-trained language models (LMs) and knowledge graphs (KGs) presents two challenges: given a QA context (question and answer choice), methods need to (i) identify relevant knowledge from large KGs, and (ii) perform joint reasoning over the QA context and KG. Here we propose a new model, QA-GNN, which addresses the above challenges through two key innovations: (i) relevance scoring, where we use LMs to estimate the importance of KG nodes relative to the given QA context, and (ii) joint reasoning, where we connect the QA context and KG to form a joint graph, and mutually update their representations through graph-based message passing. We evaluate QA-GNN on the CommonsenseQA and OpenBookQA datasets, and show its improvement over existing LM and LM+KG models, as well as its capability to perform interpretable and structured reasoning, e.g., correctly handling negation in questions. 

   more » « less
4. Integrating Heterogeneous Sources for Learned Prediction of Vehicular Data Consumption

   https://doi.org/10.1109/MDM55031.2022.00029  

   Zang, Andi; Zhu, Xiaofeng; Li, Ce; Zhou, Fan; Trajcevski, Goce (June 2022, 23rd IEEE International Conference on Mobile Data Management, MDM 2022)

   In addition to the multiple sensors to measure parameters that can be used to improve both safety and efficiency, modern vehicles also gather information about external data (e.g., traffic conditions, weather) which, if properly used, could further improve the overall trip experience. Specifically, when it comes to navigation, one source that can provide increased context awareness, especially for autonomous driving, are the High Definition (HD) maps, which have recently witnessed a tremendous growth of popularity in vehicular technology and use. As they are limited to a particular geographic area, different portions need to be downloaded (and processed) on multiple occasions throughout a given trip, along with the other data from other internal and external sources. In this paper, we provide an effective deep learning approach for the recently introduced problem of Predicting Map Data Consumption (PMDC) in the future time instants for a given trip. We propose a novel methodology that integrates multiple data sources (road network, traffic, historic trips, HD maps) and, for a given trip, enables prediction of the map data consumption. Our experimental observations demonstrate the benefits of the proposed approach over the candidate baselines. 

   more » « less
5. Pretraining Language Models with Human Preferences

   Korbak, Tomasz; Shi, Kejian; Chen, Angelica; Bhalerao, Rasika; Buckley, Christopher L.; Phang, Jason; Bowman, Samuel R.; Perez, Ethan (July 2023, International Conference on Machine Learning)

   Language models (LMs) are pretrained to imitate text from large and diverse datasets that contain content that would violate human preferences if generated by an LM: falsehoods, offensive comments, personally identifiable information, low-quality or buggy code, among others. Here, we explore alternative objectives for pretraining LMs in a way that also guides them to generate text aligned with human preferences. We benchmark five objectives for pretraining with human feedback across three tasks and study how they affect the alignment and capabilities of pretrained LMs. We find a Pareto-optimal and simple approach among those we explored: conditional training, or learning distribution over tokens conditional on their human preference scores. Conditional training reduces the rate of undesirable content by up to an order of magnitude, both when generating without a prompt and with an adversarially-chosen prompt. Moreover, conditional training maintains the downstream task performance of standard LM pretraining, both before and after task-specific finetuning. Pretraining with human feedback results in much better preference satisfaction than standard LM pretraining followed by finetuning with feedback, i.e., learning and then unlearning undesirable behavior. Our results suggest that we should move beyond imitation learning when pretraining LMs and incorporate human preferences from the start of training. 

   more » « less