Search

We have previously shown all understanding or learning are compression, under reasonable assumptions. In principle, better understanding of data should improve data compression. Traditional compression methodologies focus on encoding frequencies or some other computable properties of data. Large language models approximate the uncomputable Solomonoff distribution, opening up a whole new avenue to justify our theory. Under the new uncomputable paradigm, we present LMCompress based on the understanding of data using large models. LMCompress has significantly better lossless compression ratios than all other lossless data compression methods, doubling the compression ratios of JPEG-XL for images, FLAC for audios and H264 for videos, and tripling or quadrupling the compression ratio of bz2 for texts. The better a large model understands the data, the better LMCompress compresses.

We investigate the problem of Min-cost Perfect Matching with Delays (MPMD) in which requests are pairwise matched in an online fashion with the objective to minimize the sum of space cost and time cost. Though linear-MPMD (i.e., time cost is linear in delay) has been thoroughly studied in the literature, it does not well model impatient requests that are common in practice. Thus, we propose convex-MPMD where time cost functions are convex, capturing the situation where time cost increases faster and faster. Since the existing algorithms for linear-MPMD are not competitive any more, we devise a new deterministic algorithm for convex-MPMD problems. For a large class of convex time cost functions, our algorithm achieves a competitive ratio of $O(k)$ on any $k$-point uniform metric space, or $O(k\Delta)$ when the metric space has aspect ratio $\Delta$. Moreover, we prove lower bounds for the competitive ratio of deterministic and randomized algorithms, indicating that our deterministic algorithm is optimal. This optimality uncover a substantial difference between convex-MPMD and linear-MPMD, since linear-MPMD allows a deterministic algorithm with constant competitive ratio on any uniform metric space.

Span extraction, aiming to extract text spans (such as words or phrases) from plain texts, is a fundamental process in Information Extraction. Recent works introduce the label knowledge to enhance the text representation by formalizing the span extraction task into a question answering problem (QA Formalization), which achieves state-of-the-art performance. However, QA Formalization does not fully exploit the label knowledge and suffers from low efficiency in training/inference. To address those problems, we introduce a new paradigm to integrate label knowledge and further propose a novel model to explicitly and efficiently integrate label knowledge into text representations. Specifically, it encodes texts and label annotations independently and then integrates label knowledge into text representation with an elaborate-designed semantics fusion module. We conduct extensive experiments on three typical span extraction tasks: flat NER, nested NER, and event detection. The empirical results show that 1) our method achieves state-of-the-art performance on four benchmarks, and 2) reduces training time and inference time by 76% and 77% on average, respectively, compared with the QA Formalization paradigm. Our code and data are available at https://github.com/Akeepers/LEAR., Comment: Accepted to the main conference of EMNLP 2021 (long paper)

Motivated by scheduling in Geo-distributed data analysis, we propose a target location problem for multi-commodity flow (LoMuF for short). Given commodities to be sent from their resources, LoMuF aims at locating their targets so that the multi-commodity flow is optimized in some sense. LoMuF is a combination of two fundamental problems, namely, the facility location problem and the network flow problem. We study the hardness and algorithmic issues of the problem in various settings. The findings lie in three aspects. First, a series of NP-hardness and APX-hardness results are obtained, uncovering the inherent difficulty in solving this problem. Second, we propose an approximation algorithm for general undirected networks and an exact algorithm for undirected trees, which naturally induce efficient approximation algorithms on directed networks. Third, we observe separations between directed networks and undirected ones, indicating that imposing direction on edges makes the problem strictly harder. These results show the richness of the problem and pave the way to further studies.

A crucial assumption in most statistical learning theory is that samples are independently and identically distributed (i.i.d.). However, for many real applications, the i.i.d. assumption does not hold. We consider learning problems in which examples are dependent and their dependency relation is characterized by a graph. To establish algorithm-dependent generalization theory for learning with non-i.i.d. data, we first prove novel McDiarmid-type concentration inequalities for Lipschitz functions of graph-dependent random variables. We show that concentration relies on the forest complexity of the graph, which characterizes the strength of the dependency. We demonstrate that for many types of dependent data, the forest complexity is small and thus implies good concentration. Based on our new inequalities we are able to build stability bounds for learning from graph-dependent data., Comment: accepted as NeurIPS 2019 spotlight paper

Schedulability is a fundamental problem in real-time scheduling, but it has to be approximated due to the intrinsic computational hardness. As the most popular algorithm for deciding schedulability on multiprocess platforms, the speedup factor of partitioned-EDF is challenging to analyze and is far from been determined. Partitioned-EDF was first proposed in 2005 by Barush and Fisher [1], and was shown to have a speedup factor at most 3-1/m, meaning that if the input of sporadic tasks is feasible on m processors with speed one, partitioned-EDF will always return succeeded on m processors with speed 3-1/m. In 2011, this upper bound was improved to 2.6322-1/m by Chen and Chakraborty [2], and no more improvements have appeared ever since then. In this paper, we develop a novel method to discretize and regularize sporadic tasks, which enables us to improve, in the case of constrained deadlines, the speedup factor of partitioned-EDF to 2.5556-1/m, very close to the asymptotic lower bound 2.5 in [2].

Networked data, in which every training example involves two objects and may share some common objects with others, is used in many machine learning tasks such as learning to rank and link prediction. A challenge of learning from networked examples is that target values are not known for some pairs of objects. In this case, neither the classical i.i.d.\ assumption nor techniques based on complete U-statistics can be used. Most existing theoretical results of this problem only deal with the classical empirical risk minimization (ERM) principle that always weights every example equally, but this strategy leads to unsatisfactory bounds. We consider general weighted ERM and show new universal risk bounds for this problem. These new bounds naturally define an optimization problem which leads to appropriate weights for networked examples. Though this optimization problem is not convex in general, we devise a new fully polynomial-time approximation scheme (FPTAS) to solve it., Comment: accepted by AAAI. arXiv admin note: substantial text overlap with arXiv:math/0702683 by other authors

A tight criterion under which the abstract version Lov\'asz Local Lemma (abstract-LLL) holds was given by Shearer decades ago. However, little is known about that of the variable version LLL (variable-LLL) where events are generated by independent random variables, though this model of events is applicable to almost all applications of LLL. We introduce a necessary and sufficient criterion for variable-LLL, in terms of the probabilities of the events and the event-variable graph specifying the dependency among the events. Based on this new criterion, we obtain boundaries for two families of event-variable graphs, namely, cyclic and treelike bigraphs. These are the first two non-trivial cases where the variable-LLL boundary is fully determined. As a byproduct, we also provide a universal constructive method to find a set of events whose union has the maximum probability, given the probability vector and the event-variable graph. Though it is #P-hard in general to determine variable-LLL boundaries, we can to some extent decide whether a gap exists between a variable-LLL boundary and the corresponding abstract-LLL boundary. In particular, we show that the gap existence can be decided without solving Shearer's conditions or checking our variable-LLL criterion. Equipped with this powerful theorem, we show that there is no gap if the base graph of the event-variable graph is a tree, while gap appears if the base graph has an induced cycle of length at least 4. The problem is almost completely solved except when the base graph has only 3-cliques, in which case we also get partial solutions. A set of reduction rules are established that facilitate to infer gap existence of an event-variable graph from known ones. As an application, various event-variable graphs, in particular combinatorial ones, are shown to be gapful/gapless., Comment: Part of the work has been published at FOCS2017

Borgs et al. [2016] investigated essential requirements for communities in preference networks. They defined six axioms on community functions, i.e., community detection rules. Though having elegant properties, the practicality of this axiom system is compromised by the intractability of checking two critical axioms, so no nontrivial consistent community function was reported inBorgs et al. [2016] By adapting the two axioms in a natural way, we propose two new axioms that are efficiently-checkable. We show that most of the desirable properties of the original axiom system are preserved. More importantly, the new axioms provide a general approach to constructing consistent community functions. We further find a natural consistent community function that is also enumerable and samplable, answering an open problem in the literature., Comment: ICDM'16

In this paper we investigate the top-$k$-selection problem, i.e. determine the largest, second largest, ..., and the $k$-th largest elements, in the dynamic data model. In this model the order of elements evolves dynamically over time. In each time step the algorithm can only probe the changes of data by comparing a pair of elements. Previously only two special cases were studied[2]: finding the largest element and the median; and sorting all elements. This paper systematically deals with $k\in [n]$ and solves the problem almost completely. Specifically, we identify a critical point $k^*$ such that the top-$k$-selection problem can be solved error-free with probability $1-o(1)$ if and only if $k=o(k^*)$. A lower bound of the error when $k=\Omega(k^*)$ is also determined, which actually is tight under some condition. On the other hand, it is shown that the top-$k$-set problem, which means finding the largest $k$ elements without sorting them, can be solved error-free for all $k\in [n]$. Additionally, we extend the dynamic data model and show that most of these results still hold., Comment: 23 pages, 2 figures