Large language models have demonstrated strong capabilities to learn in-context, where exemplar input-output pairings are appended to the prompt for demonstration. However, existing work has demonstrated the ability of models to learn lexical and label biases in-context, which negatively impacts both performance and robustness of models. The impact of other statistical data biases remains under-explored, which this work aims to address. We specifically investigate the impact of length biases on in-context learning. We demonstrate that models do learn length biases in the context window for their predictions, and further empirically analyze the factors that modulate the level of bias exhibited by the model. In addition, we show that learning length information in-context can be used to counter the length bias that has been encoded in models (e.g., via fine-tuning). This reveals the power of in-context learning in debiasing model prediction behaviors without the need for costly parameter updates.

A major challenge for transformers is generalizing to sequences longer than those observed during training. While previous works have empirically shown that transformers can either succeed or fail at length generalization depending on the task, theoretical understanding of this phenomenon remains limited. In this work, we introduce a rigorous theoretical framework to analyze length generalization in causal transformers with learnable absolute positional encodings. In particular, we characterize those functions that are identifiable in the limit from sufficiently long inputs with absolute positional encodings under an idealized inference scheme using a norm-based regularizer. This enables us to prove the possibility of length generalization for a rich family of problems. We experimentally validate the theory as a predictor of success and failure of length generalization across a range of algorithmic and formal language tasks. Our theory not only explains a broad set of empirical observations but also opens the way to provably predicting length generalization capabilities in transformers.

In sparse linear bandits, a learning agent sequentially selects an action and receive reward feedback, and the reward function depends linearly on a few coordinates of the covariates of the actions. This has applications in many real-world sequential decision making problems. In this paper, we propose a simple and computationally efficient sparse linear estimation method called PopArt that enjoys a tighter $\ell_1$ recovery guarantee compared to Lasso (Tibshirani, 1996) in many problems. Our bound naturally motivates an experimental design criterion that is convex and thus computationally efficient to solve. Based on our novel estimator and design criterion, we derive sparse linear bandit algorithms that enjoy improved regret upper bounds upon the state of the art (Hao et al., 2020), especially w.r.t. the geometry of the given action set. Finally, we prove a matching lower bound for sparse linear bandits in the data-poor regime, which closes the gap between upper and lower bounds in prior work.

Pairwise measures of dependence are a common tool to map data in the early stages of analysis with several modern examples based on maximized partitions of the pairwise sample space. Following a short survey of modern measures of dependence, we introduce a new measure which recursively splits the ranks of a pair of variables to partition the sample space and computes the $\chi^2$ statistic on the resulting bins. Splitting logic is detailed for splits maximizing a score function and randomly selected splits. Simulations indicate that random splitting produces a statistic conservatively approximated by the $\chi^2$ distribution without a loss of power to detect numerous different data patterns compared to maximized binning. Though it seems to add no power to detect dependence, maximized recursive binning is shown to produce a natural visualization of the data and the measure. Applying maximized recursive rank binning to S&P 500 constituent data suggests the automatic detection of tail dependence.

We report and compare the performance of different learning algorithms based on data from cortical recordings. The task is to predict the orientation of visual stimuli from the activity of a population of simultaneously recorded neurons. We compare several ways of improving the coding of the input (i.e., the spike data) as well as of the output (i.e., the orientation), and report the results obtained using different kernel algorithms.

We report and compare the performance of different learning algorithms based on data from cortical recordings. The task is to predict the orientation of visual stimuli from the activity of a population of simultaneously recorded neurons. We compare several ways of improving the coding of the input (i.e., the spike data) as well as of the output (i.e., the orientation), and report the results obtained using different kernel algorithms.