Zachary F. Fisher

Santiago Ontanon, Joshua Ainslie, Vaclav Cvicek et al. · deepmind

Machine learning models such as Transformers or LSTMs struggle with tasks that are compositional in nature such as those involving reasoning/inference. Although many datasets exist to evaluate compositional generalization, when it comes to evaluating inference abilities, options are more limited. This paper presents LogicInference, a new dataset to evaluate the ability of models to perform logical inference. The dataset focuses on inference using propositional logic and a small subset of first-order logic, represented both in semi-formal logical notation, as well as in natural language. We also report initial results using a collection of machine learning models to establish an initial baseline in this dataset.

Renat Aksitov, Sobhan Miryoosefi, Zonglin Li et al. · deepmind

Answering complex natural language questions often necessitates multi-step reasoning and integrating external information. Several systems have combined knowledge retrieval with a large language model (LLM) to answer such questions. These systems, however, suffer from various failure cases, and we cannot directly train them end-to-end to fix such failures, as interaction with external knowledge is non-differentiable. To address these deficiencies, we define a ReAct-style LLM agent with the ability to reason and act upon external knowledge. We further refine the agent through a ReST-like method that iteratively trains on previous trajectories, employing growing-batch reinforcement learning with AI feedback for continuous self-improvement and self-distillation. Starting from a prompted large model and after just two iterations of the algorithm, we can produce a fine-tuned small model that achieves comparable performance on challenging compositional question-answering benchmarks with two orders of magnitude fewer parameters.

Santiago Ontañón, Joshua Ainslie, Vaclav Cvicek et al.

Several studies have reported the inability of Transformer models to generalize compositionally, a key type of generalization in many NLP tasks such as semantic parsing. In this paper we explore the design space of Transformer models showing that the inductive biases given to the model by several design decisions significantly impact compositional generalization. Through this exploration, we identified Transformer configurations that generalize compositionally significantly better than previously reported in the literature in a diverse set of compositional tasks, and that achieve state-of-the-art results in a semantic parsing compositional generalization benchmark (COGS), and a string edit operation composition benchmark (PCFG).

Zachary F. Fisher, Younghoon Kim, Barbara Fredrickson et al.

Intensive Longitudinal Data (ILD) is increasingly available to social and behavioral scientists. With this increased availability come new opportunities for modeling and predicting complex biological, behavioral, and physiological phenomena. Despite these new opportunities psychological researchers have not taken full advantage of promising opportunities inherent to this data, the potential to forecast psychological processes at the individual level. To address this gap in the literature we present a novel modeling framework that addresses a number of topical challenges and open questions in the psychological literature on modeling dynamic processes. First, how can we model and forecast ILD when the length of individual time series and the number of variables collected are roughly equivalent, or when time series lengths are shorter than what is typically required for time series analyses? Second, how can we best take advantage of the cross-sectional (between-person) information inherent to most ILD scenarios while acknowledging individuals differ both quantitatively (e.g. in parameter magnitude) and qualitatively (e.g. in structural dynamics)? Despite the acknowledged between-person heterogeneity in many psychological processes is it possible to leverage group-level information to support improved forecasting at the individual level? In the remainder of the manuscript, we attempt to address these and other pressing questions relevant to the forecasting of multiple-subject ILD.