arXiv subjects

Lusheng Wang

Publications and source records attributed to Lusheng Wang.

14 recordsLinked to original sources

arXiv2026

Towards general embodied intelligence: integrating large language models, knowledge bases, and reasoning capabilities to build the next generation of AI agents

The convergence of large language models (LLMs), structured knowledge bases (KBs), and reasoning ability (RA) presents a promising trajectory toward general embodied intelligence (GEI). This paper reviews the evolution of LLM-centered intelligent systems, emphasising their integration with knowledge representation, logical reasoning, and physical embodiment. We analyse LLM architectures, pre-training methods, and inference mechanisms, along with their interaction with external knowledge sources and structured reasoning frameworks. Furthermore, we examine embodied intelligence (EI) paradigms wherein agents learn and act in physical environments. To synthesise these dimensions, we present a conceptual framework that illustrates the synergy among LLMs, KBs, RA, and embodiment, serving as a guiding model for perception, reasoning, and action rather than an implemented engineering architecture. To advance toward GEI, we identify five key challenges: efficient LLM deployment, closed-loop knowledge integration, hybrid symbolic-neural reasoning, perception-action grounding, and continual learning. This survey provides a comprehensive roadmap for developing adaptive, multimodal agents capable of operating in complex, dynamic settings.

cs.AI↗

arXiv2026

Branch-JEPA: Finite-Support Predictive Distributions for JEPA World Models

Joint-embedding predictive architectures (JEPAs) learn dynamics by predicting future observations in representation space. Yet most JEPA world models return one latent successor, even when hidden intent, partial observation, or stochastic dynamics make several futures plausible. We introduce Branch-JEPA, which replaces this point-valued transition with a context-weighted finite set of latent successors. Every branch is decoded independently, and the complete set is retained at inference. The architecture supports two complementary training regimes: specialization for recovering separated successors and full-set Energy-Score training for distributional fidelity. In a locked five-seed evaluation on the Argoverse~2 official validation split, full-set training improves trajectory Energy Score by $5.8$--$6.5\%$ and probability-weighted trajectory distance by $9.3$--$10.4\%$ over matched-$K{=}6$ assignment and transport objectives, while retaining $5.36$ endpoint-deduplicated effective branches. In a parameter-exact official-validation comparison, latent branching retains $10.3\%$ more effective modes and improves Energy Score, expected ADE, and Brier in all five paired seeds over branching only at the output decoder; every paired 95\% interval excludes zero. In an OGBench graph audit, Branch-JEPA increases teleport verified-route existence to $19.2\%$ versus $3.9\%$ for the MDN. Its raw-support advantage also persists with 29-D state and RGB observations. Together, latent branching preserves more distinct futures, while full-set scoring improves the quality of the resulting predictive distribution.

cs.AI↗

arXiv2024

Deduction, Constrained Zero Forcing, and Constrained Searching

Deduction is a recently introduced graph searching process in which searchers clear the vertex set of a graph with one move each, with each searcher's movement determined by which of its neighbors are protected by other searchers. In this paper, we show that the minimum number of searchers required to clear the graph is the same in deduction as in constrained versions of other previously studied graph processes, namely zero forcing and fast-mixed search. We give a structural characterization, new bounds and a spectrum result on the number of searchers required. We consider the complexity of computing this parameter, giving an NP-completeness result for arbitrary graphs, and exhibiting families of graphs for which the parameter can be computed in polynomial time. We also describe properties of the deduction process related to the timing of searcher movement and the success of terminal layouts.

math.CO↗

arXiv2024

Approximately covering vertices by order-$5$ or longer paths

This paper studies $MPC^{5+}_v$, which is to cover as many vertices as possible in a given graph $G=(V,E)$ by vertex-disjoint $5^+$-paths (i.e., paths each with at least five vertices). $MPC^{5+}_v$ is NP-hard and admits an existing local-search-based approximation algorithm which achieves a ratio of $\frac {19}7\approx 2.714$ and runs in $O(|V|^6)$ time. In this paper, we present a new approximation algorithm for $MPC^{5+}_v$ which achieves a ratio of $2.511$ and runs in $O(|V|^{2.5} |E|^2)$ time. Unlike the previous algorithm, the new algorithm is based on maximum matching, maximum path-cycle cover, and recursion.

cs.DS↗

arXiv2023

Randomized algorithms for fully online multiprocessor scheduling with testing

We contribute the first randomized algorithm that is an integration of arbitrarily many deterministic algorithms for the fully online multiprocessor scheduling with testing problem. When there are two machines, we show that with two component algorithms its expected competitive ratio is already strictly smaller than the best proven deterministic competitive ratio lower bound. Such algorithmic results are rarely seen in the literature. Multiprocessor scheduling is one of the first combinatorial optimization problems that have received numerous studies. Recently, several research groups examined its testing variant, in which each job $J_j$ arrives with an upper bound $u_j$ on the processing time and a testing operation of length $t_j$; one can choose to execute $J_j$ for $u_j$ time, or to test $J_j$ for $t_j$ time to obtain the exact processing time $p_j$ followed by immediately executing the job for $p_j$ time. Our target problem is the fully online version, in which the jobs arrive in sequence so that the testing decision needs to be made at the job arrival as well as the designated machine. We propose an expected $(\sqrt{φ+ 3} + 1) (\approx 3.1490)$-competitive randomized algorithm as a non-uniform probability distribution over arbitrarily many deterministic algorithms, where $φ= \frac {\sqrt{5} + 1}2$ is the Golden ratio. When there are two machines, we show that our randomized algorithm based on two deterministic algorithms is already expected $\frac {3 φ+ 3 \sqrt{13 - 7φ}}4 (\approx 2.1839)$-competitive. Besides, we use Yao's principle to prove lower bounds of $1.6682$ and $1.6522$ on the expected competitive ratio for any randomized algorithm at the presence of at least three machines and only two machines, respectively, and prove a lower bound of $2.2117$ on the competitive ratio for any deterministic algorithm when there are only two machines.

cs.DS↗

arXiv2023

Men Can't Always be Transformed into Mice: Decision Algorithms and Complexity for Sorting by Symmetric Reversals

Sorting a permutation by reversals is a famous problem in genome rearrangements. Since 1997, quite some biological evidence were found that in many genomes the reversed regions are usually flanked by a pair of inverted repeats. This type of reversals are called symmetric reversals, which, unfortunately, were largely ignored until recently. In this paper, we investigate the problem of sorting by symmetric reversals, which requires a series of symmetric reversals to transform one chromosome $A$ into the another chromosome $B$. The decision problem of sorting by symmetric reversals is referred to as {\em SSR} (when the input chromosomes $A$ and $B$ are given, we use {\em SSR(A,B)}) and the corresponding optimization version (i.e., when the answer for {\em SSR(A,B)} is yes, using the minimum number of symmetric reversals to convert $A$ to $B$), is referred to as {\em SMSR(A,B)}. The main results of this paper are summarized as follows, where the input is a pair of chromosomes $A$ and $B$ with $n$ repeats. (1) We present an $O(n^2)$ time algorithm to solve the decision problem {\em SSR(A,B)}, i.e., determine whether a chromosome $A$ can be transformed into $B$ by a series of symmetric reversals. (2) We design an $O(n^2)$ time algorithm for a special 2-balanced case of {\em SMSR(A,B)}, where chromosomes $A$ and $B$ both have duplication number 2 and every repeat appears twice in different orientations in $A$ and $B$. (3) We show that SMSR is NP-hard even if the duplication number of the input chromosomes are at most 2, hence showing that the above positive optimization result is the best possible. As a by-product, we show that the \emph{minimum Steiner tree} problem on \emph{circle graphs} is NP-hard, settling the complexity status of a 38-year old open problem.

cs.DS↗

arXiv2017

A Cubic-Time 2-Approximation Algorithm for rSPR Distance

Due to hybridization events in evolution, studying two different genes of a set of species may yield two related but different phylogenetic trees for the set of species. In this case, we want to measure the dissimilarity of the two trees. The rooted subtree prune and regraft (rSPR) distance of the two trees has been used for this purpose. The problem of computing the rSPR distance of two given trees has many applications but is unfortunately NP-hard. The previously best approximation algorithm for rSPR distance achieves a ratio of 2.5 and it was open whether a better approximation algorithm for rSPR distance exists. In this paper, we answer this question in the affirmative by presenting a cubic-time approximation algorithm for rSPR distance that achieves a ratio of 2. Our algorithm is based on the new notion of key and a number of new structural lemmas. The algorithm is fairly simple and the proof of its correctness is intuitively understandable albeit complicated.

cs.DS↗

arXiv2017

Approximation algorithms for the maximum weight internal spanning tree problem

Given a vertex-weighted connected graph $G = (V, E)$, the maximum weight internal spanning tree (MwIST for short) problem asks for a spanning tree $T$ of $G$ such that the total weight of the internal vertices in $T$ is maximized. The un-weighted variant, denoted as MIST, is NP-hard and APX-hard, and the currently best approximation algorithm has a proven performance ratio $13/17$. The currently best approximation algorithm for MwIST only has a performance ratio $1/3 - ε$, for any $ε> 0$. In this paper, we present a simple algorithm based on a novel relationship between MwIST and the maximum weight matching, and show that it achieves a better approximation ratio of $1/2$. When restricted to claw-free graphs, a special case been previously studied, we design a $7/12$-approximation algorithm.

cs.DS↗

arXiv2016

Core-genome scaffold comparison reveals the prevalence that inversion events are associated with pairs of inverted repeats

Motivation: Genome rearrangement plays an important role in evolutionary biology and has profound impacts on phenotype in organisms ranging from microbes to humans. The mechanisms for genome rearrangement events remain unclear. Lots of comparisons have been conducted among different species. To reveal the mechanisms for rearrangement events, comparison of different individuals/strains within the same species or genus (pan-genomes) is more helpful since they are much closer to each other. Results: We study the mechanism for inversion events via core-genome scaffold comparison of different strains within the same species. We focus on two kinds of bacteria, Pseudomonas aeruginosa and Escherichia coli, and investigate the inversion events among different strains of the same specie. We find an interesting phenomenon that long (larger than 10,000 bp) inversion regions are flanked by a pair of Inverted Repeats (IRs) (with lengths ranging from 385 bp to 27476 bp) which are often Insertion Sequences (ISs).This mechanism can also explain why the breakpoint reuses for inversion events happen. We study the prevalence of the phenomenon and find that it is a major mechanism for inversions. The other observation is that for different rearrangement events such as transposition and inverted block interchange, the two ends of the swapped regions are also associated with repeats so that after the rearrangement operations the two ends of the swapped regions remain unchanged. To our knowledge, this is the first time such a phenomenon is reported for transposition event.

q-bio.GN↗

arXiv2016

An Approximation Algorithm for Maximum Internal Spanning Tree

Given a graph G, the {\em maximum internal spanning tree problem} (MIST for short) asks for computing a spanning tree T of G such that the number of internal vertices in T is maximized. MIST has possible applications in the design of cost-efficient communication networks and water supply networks and hence has been extensively studied in the literature. MIST is NP-hard and hence a number of polynomial-time approximation algorithms have been designed for MIST in the literature. The previously best polynomial-time approximation algorithm for MIST achieves a ratio of 3/4. In this paper, we first design a simpler algorithm that achieves the same ratio and the same time complexity as the previous best. We then refine the algorithm into a new approximation algorithm that achieves a better ratio (namely, 13/17) with the same time complexity. Our new algorithm explores much deeper structure of the problem than the previous best. The discovered structure may be used to design even better approximation or parameterized algorithms for the problem in the future.

cs.DS↗

arXiv2014

To Achieve Maximal Throughputs in CSMA Wireless Networks Through Offered_load Control

This paper studies how to achieve the maximal link throughputs in a CSMA wireless network through offered-load control. First, we propose an analytical model, contention-graph-combination (CGC), to describe the relationship between the offered-load and the output link throughputs of an unsaturated CSMA network. Based on CGC, we then formulate a linear optimization model to improve the aggregate link throughput through properly setting the occurrence probabilities of each sub-network, based on which we can obtain the optimal offered-load of each link. Simulation results bore out the accuracy of our CGC analysis and the maximal link throughputs can be closely achieved. Different from prior work in which CSMA protocol parameters are adaptively adjusted to achieve better performance, in this paper we propose to achieve maximal link throughputs by adjusting the rates of the traffic pumped into the source nodes of links, which runs in a software manner and is more practical to implement in real networks.

cs.NI↗

arXiv2013

An approximation algorithm for the Bandpass-2 problem

The general Bandpass-$B$ problem is NP-hard and can be approximated by a reduction into the weighted $B$-set packing problem, with a worst case performance ratio of $O(B^2)$. When $B = 2$, a maximum weight matching gives a 2-approximation to the problem. In this paper, we call the Bandpass-2 problem simply the Bandpass problem. The Bandpass problem can be viewed as a variation of the maximum traveling salesman problem, in which the edge weights are dynamic rather than given at the front. We present a ${426}{227}$-approximation algorithm for the problem. Such an improved approximation is built on an intrinsic structural property proven for the optimal solution and several novel schemes to partition a $b$-matching into desired matchings.

cs.DS↗

arXiv2013

Global Existence and Decay of Solutions to the Fokker-Planck-Boltzmann Equation

The Cauchy problem to the Fokker-Planck-Boltzmann equation under Grad's angular cut-off assumption is investigated. When the initial data is a small perturbation of an equilibrium state, global existence and optimal temporal decay estimates of classical solutions are established. Our analysis is based on the coercivity of the Fokker-Planck operator and an elementary weighted energy method.

math.AP↗

arXiv2000

On The Closest String and Substring Problems

The problem of finding a center string that is `close' to every given string arises and has many applications in computational biology and coding theory. This problem has two versions: the Closest String problem and the Closest Substring problem. Assume that we are given a set of strings ${\cal S}=\{s_1, s_2, ..., s_n\}$ of strings, say, each of length $m$. The Closest String problem asks for the smallest $d$ and a string $s$ of length $m$ which is within Hamming distance $d$ to each $s_i\in {\cal S}$. This problem comes from coding theory when we are looking for a code not too far away from a given set of codes. The problem is NP-hard. Berman et al give a polynomial time algorithm for constant $d$. For super-logarithmic $d$, Ben-Dor et al give an efficient approximation algorithm using linear program relaxation technique. The best polynomial time approximation has ratio 4/3 for all $d$ given by Lanctot et al and Gasieniec et al. The Closest Substring problem looks for a string $t$ which is within Hamming distance $d$ away from a substring of each $s_i$. This problem only has a $2- \frac{2}{2|Σ|+1}$ approximation algorithm previously Lanctot et al and is much more elusive than the Closest String problem, but it has many applications in finding conserved regions, genetic drug target identification, and genetic probes in molecular biology. Whether there are efficient approximation algorithms for both problems are major open questions in this area. We present two polynomial time approxmation algorithms with approximation ratio $1+ ε$ for any small $ε$ to settle both questions.

cs.CE↗