Large language models (LLMs) have shown promising efficacy across various tasks, becoming powerful tools in numerous aspects of human life. However, Transformer-based LLMs suffer a performance degradation when modeling long-term contexts due to they discard some information to reduce computational overhead. In this work, we propose a simple yet effective method to enable LLMs to take a deep breath, encouraging them to summarize information contained within discrete text chunks. Specifically, we segment the text into multiple chunks and insert special token <SR> at the end of each chunk. We then modify the attention mask to integrate the chunk's information into the corresponding <SR> token. This facilitates LLMs to interpret information not only from historical individual tokens but also from the <SR> token, aggregating the chunk's semantic information. Experiments on language modeling and out-of-domain downstream tasks validate the superiority of our approach.
This paper addresses the challenge of reconstructing surfaces from sparse view inputs, where ambiguity and occlusions due to missing information pose significant hurdles. We present a novel approach, named EpiS, that incorporates Epipolar information into the reconstruction process. Existing methods in sparse-view neural surface learning have mainly focused on mean and variance considerations using cost volumes for feature extraction. In contrast, our method aggregates coarse information from the cost volume into Epipolar features extracted from multiple source views, enabling the generation of fine-grained Signal Distance Function (SDF)-aware features. Additionally, we employ an attention mechanism along the line dimension to facilitate feature fusion based on the SDF feature. Furthermore, to address the information gaps in sparse conditions, we integrate depth information from monocular depth estimation using global and local regularization techniques. The global regularization utilizes a triplet loss function, while the local regularization employs a derivative loss function. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods, especially in cases with sparse and generalizable conditions.
When optimizing machine learning models, there are various scenarios where gradient computations are challenging or even infeasible. Furthermore, in reinforcement learning (RL), preference-based RL that only compares between options has wide applications, including reinforcement learning with human feedback in large language models. In this paper, we systematically study optimization of a smooth function $f\colon\mathbb{R}^n\to\mathbb{R}$ only assuming an oracle that compares function values at two points and tells which is larger. When $f$ is convex, we give two algorithms using $\tilde{O}(n/\epsilon)$ and $\tilde{O}(n^{2})$ comparison queries to find an $\epsilon$-optimal solution, respectively. When $f$ is nonconvex, our algorithm uses $\tilde{O}(n/\epsilon^2)$ comparison queries to find an $\epsilon$-approximate stationary point. All these results match the best-known zeroth-order algorithms with function evaluation queries in $n$ dependence, thus suggest that \emph{comparisons are all you need for optimizing smooth functions using derivative-free methods}. In addition, we also give an algorithm for escaping saddle points and reaching an $\epsilon$-second order stationary point of a nonconvex $f$, using $\tilde{O}(n^{1.5}/\epsilon^{2.5})$ comparison queries.
The sim-to-real gap, which represents the disparity between training and testing environments, poses a significant challenge in reinforcement learning (RL). A promising approach to addressing this challenge is distributionally robust RL, often framed as a robust Markov decision process (RMDP). In this framework, the objective is to find a robust policy that achieves good performance under the worst-case scenario among all environments within a pre-specified uncertainty set centered around the training environment. Unlike previous work, which relies on a generative model or a pre-collected offline dataset enjoying good coverage of the deployment environment, we tackle robust RL via interactive data collection, where the learner interacts with the training environment only and refines the policy through trial and error. In this robust RL paradigm, two main challenges emerge: managing distributional robustness while striking a balance between exploration and exploitation during data collection. Initially, we establish that sample-efficient learning without additional assumptions is unattainable owing to the curse of support shift; i.e., the potential disjointedness of the distributional supports between the training and testing environments. To circumvent such a hardness result, we introduce the vanishing minimal value assumption to RMDPs with a total-variation (TV) distance robust set, postulating that the minimal value of the optimal robust value function is zero. We prove that such an assumption effectively eliminates the support shift issue for RMDPs with a TV distance robust set, and present an algorithm with a provable sample complexity guarantee. Our work makes the initial step to uncovering the inherent difficulty of robust RL via interactive data collection and sufficient conditions for designing a sample-efficient algorithm accompanied by sharp sample complexity analysis.
Time series modeling presents unique challenges due to autocorrelation in both historical data and future sequences. While current research predominantly addresses autocorrelation within historical data, the correlations among future labels are often overlooked. Specifically, modern forecasting models primarily adhere to the Direct Forecast (DF) paradigm, generating multi-step forecasts independently and disregarding label autocorrelation over time. In this work, we demonstrate that the learning objective of DF is biased in the presence of label autocorrelation. To address this issue, we propose the Frequency-enhanced Direct Forecast (FreDF), which mitigates label autocorrelation by learning to forecast in the frequency domain, thereby reducing estimation bias. Our experiments show that FreDF significantly outperforms existing state-of-the-art methods and is compatible with a variety of forecast models. Code is available at https://github.com/Master-PLC/FreDF.
Signatures are iterated path integrals of continuous and discrete-time processes, and their universal nonlinearity linearizes the problem of feature selection in time series data analysis. This paper studies the consistency of signature using Lasso regression, both theoretically and numerically. We establish conditions under which the Lasso regression is consistent both asymptotically and in finite sample. Furthermore, we show that the Lasso regression is more consistent with the It\^o signature for time series and processes that are closer to the Brownian motion and with weaker inter-dimensional correlations, while it is more consistent with the Stratonovich signature for mean-reverting time series and processes. We demonstrate that signature can be applied to learn nonlinear functions and option prices with high accuracy, and the performance depends on properties of the underlying process and the choice of the signature.
Semiparametric language models (LMs) have shown promise in continuously learning from new text data by combining a parameterized neural LM with a growable non-parametric memory for memorizing new content. However, conventional semiparametric LMs will finally become prohibitive for computing and storing if they are applied to continual learning over streaming data, because the non-parametric memory grows linearly with the amount of data they learn from over time. To address the issue of scalability, we present a simple and intuitive approach called Selective Memorization (SeMem), which only memorizes difficult samples that the model is likely to struggle with. We demonstrate that SeMem improves the scalability of semiparametric LMs for continual learning over streaming data in two ways: (1) data-wise scalability: as the model becomes stronger through continual learning, it will encounter fewer difficult cases that need to be memorized, causing the growth of the non-parametric memory to slow down over time rather than growing at a linear rate with the size of training data; (2) model-wise scalability: SeMem allows a larger model to memorize fewer samples than its smaller counterpart because it is rarer for a larger model to encounter incomprehensible cases, resulting in a non-parametric memory that does not scale linearly with model size. We conduct extensive experiments in language modeling and downstream tasks to test SeMem's results, showing SeMem enables a semiparametric LM to be a scalable continual learner with little forgetting.
We study the statistical inference of nonlinear stochastic approximation algorithms utilizing a single trajectory of Markovian data. Our methodology has practical applications in various scenarios, such as Stochastic Gradient Descent (SGD) on autoregressive data and asynchronous Q-Learning. By utilizing the standard stochastic approximation (SA) framework to estimate the target parameter, we establish a functional central limit theorem for its partial-sum process, $\boldsymbol{\phi}_T$. To further support this theory, we provide a matching semiparametric efficient lower bound and a non-asymptotic upper bound on its weak convergence, measured in the L\'evy-Prokhorov metric. This functional central limit theorem forms the basis for our inference method. By selecting any continuous scale-invariant functional $f$, the asymptotic pivotal statistic $f(\boldsymbol{\phi}_T)$ becomes accessible, allowing us to construct an asymptotically valid confidence interval. We analyze the rejection probability of a family of functionals $f_m$, indexed by $m \in \mathbb{N}$, through theoretical and numerical means. The simulation results demonstrate the validity and efficiency of our method.
The 20 Questions (Q20) game is a well known game which encourages deductive reasoning and creativity. In the game, the answerer first thinks of an object such as a famous person or a kind of animal. Then the questioner tries to guess the object by asking 20 questions. In a Q20 game system, the user is considered as the answerer while the system itself acts as the questioner which requires a good strategy of question selection to figure out the correct object and win the game. However, the optimal policy of question selection is hard to be derived due to the complexity and volatility of the game environment. In this paper, we propose a novel policy-based Reinforcement Learning (RL) method, which enables the questioner agent to learn the optimal policy of question selection through continuous interactions with users. To facilitate training, we also propose to use a reward network to estimate the more informative reward. Compared to previous methods, our RL method is robust to noisy answers and does not rely on the Knowledge Base of objects. Experimental results show that our RL method clearly outperforms an entropy-based engineering system and has competitive performance in a noisy-free simulation environment.