arXiv subjects

Wing-Kin Sung

Publications and source records attributed to Wing-Kin Sung.

13 recordsLinked to original sources

arXiv2020

Studying The Effect of MIL Pooling Filters on MIL Tasks

There are different multiple instance learning (MIL) pooling filters used in MIL models. In this paper, we study the effect of different MIL pooling filters on the performance of MIL models in real world MIL tasks. We designed a neural network based MIL framework with 5 different MIL pooling filters: `max', `mean', `attention', `distribution' and `distribution with attention'. We also formulated 5 different MIL tasks on a real world lymph node metastases dataset. We found that the performance of our framework in a task is different for different filters. We also observed that the performances of the five pooling filters are also different from task to task. Hence, the selection of a correct MIL pooling filter for each MIL task is crucial for better performance. Furthermore, we noticed that models with `distribution' and `distribution with attention' pooling filters consistently perform well in almost all of the tasks. We attribute this phenomena to the amount of information captured by `distribution' based pooling filters. While point estimate based pooling filters, like `max' and `mean', produce point estimates of distributions, `distribution' based pooling filters capture the full information in distributions. Lastly, we compared the performance of our neural network model with `distribution' pooling filter with the performance of the best MIL methods in the literature on classical MIL datasets and our model outperformed the others.

cs.CV↗

arXiv2020

Weakly Supervised Clustering by Exploiting Unique Class Count

A weakly supervised learning based clustering framework is proposed in this paper. As the core of this framework, we introduce a novel multiple instance learning task based on a bag level label called unique class count ($ucc$), which is the number of unique classes among all instances inside the bag. In this task, no annotations on individual instances inside the bag are needed during training of the models. We mathematically prove that with a perfect $ucc$ classifier, perfect clustering of individual instances inside the bags is possible even when no annotations on individual instances are given during training. We have constructed a neural network based $ucc$ classifier and experimentally shown that the clustering performance of our framework with our weakly supervised $ucc$ classifier is comparable to that of fully supervised learning models where labels for all instances are known. Furthermore, we have tested the applicability of our framework to a real world task of semantic segmentation of breast cancer metastases in histological lymph node sections and shown that the performance of our weakly supervised framework is comparable to the performance of a fully supervised Unet model.

cs.CV↗

arXiv2017

A Faster Construction of Greedy Consensus Trees

A consensus tree is a phylogenetic tree that captures the similarity between a set of conflicting phylogenetic trees. The problem of computing a consensus tree is a major step in phylogenetic tree reconstruction. It also finds applications in predicting a species tree from a set of gene trees. This paper focuses on two of the most well-known and widely used oconsensus tree methods: the greedy consensus tree and the frequency difference consensus tree. Given $k$ conflicting trees each with $n$ leaves, the previous fastest algorithms for these problems were $O(k n^2)$ for the greedy consensus tree [J. ACM 2016] and $\tilde O(\min \{ k n^2, k^2n\})$ for the frequency difference consensus tree [ACM TCBB 2016]. We improve these running times to $\tilde O(k n^{1.5})$ and $\tilde O(k n)$ respectively.

cs.DS↗

arXiv2017

Faster algorithms for 1-mappability of a sequence

In the k-mappability problem, we are given a string x of length n and integers m and k, and we are asked to count, for each length-m factor y of x, the number of other factors of length m of x that are at Hamming distance at most k from y. We focus here on the version of the problem where k = 1. The fastest known algorithm for k = 1 requires time O(mn log n/ log log n) and space O(n). We present two algorithms that require worst-case time O(mn) and O(n log^2 n), respectively, and space O(n), thus greatly improving the state of the art. Moreover, we present an algorithm that requires average-case time and space O(n) for integer alphabets if m = Ω(log n/ log σ), where σ is the alphabet size.

cs.DS↗

arXiv2013

Algorithms for the Majority Rule (+) Consensus Tree and the Frequency Difference Consensus Tree

This paper presents two new deterministic algorithms for constructing consensus trees. Given an input of k phylogenetic trees with identical leaf label sets and n leaves each, the first algorithm constructs the majority rule (+) consensus tree in O(kn) time, which is optimal since the input size is Omega(kn), and the second one constructs the frequency difference consensus tree in min(O(kn^2), O(kn (k+log^2 n))) time.

cs.DS↗

arXiv2012

CRAM: Compressed Random Access Memory

We present a new data structure called the \emph{Compressed Random Access Memory} (CRAM) that can store a dynamic string $T$ of characters, e.g., representing the memory of a computer, in compressed form while achieving asymptotically almost-optimal bounds (in terms of empirical entropy) on the compression ratio. It allows short substrings of $T$ to be decompressed and retrieved efficiently and, significantly, characters at arbitrary positions of $T$ to be modified quickly during execution \emph{without decompressing the entire string}. This can be regarded as a new type of data compression that can update a compressed file directly. Moreover, at the cost of slightly increasing the time spent per operation, the CRAM can be extended to also support insertions and deletions. Our key observation that the empirical entropy of a string does not change much after a small change to the string, as well as our simple yet efficient method for maintaining an array of variable-length blocks under length modifications, may be useful for many other applications as well.

cs.DS↗

arXiv2008

Fixed Parameter Polynomial Time Algorithms for Maximum Agreement and Compatible Supertrees

Consider a set of labels $L$ and a set of trees ${\mathcal T} = \{{\mathcal T}^{(1), {\mathcal T}^{(2), ..., {\mathcal T}^{(k) \$ where each tree ${\mathcal T}^{(i)$ is distinctly leaf-labeled by some subset of $L$. One fundamental problem is to find the biggest tree (denoted as supertree) to represent $\mathcal T}$ which minimizes the disagreements with the trees in ${\mathcal T}$ under certain criteria. This problem finds applications in phylogenetics, database, and data mining. In this paper, we focus on two particular supertree problems, namely, the maximum agreement supertree problem (MASP) and the maximum compatible supertree problem (MCSP). These two problems are known to be NP-hard for $k \geq 3$. This paper gives the first polynomial time algorithms for both MASP and MCSP when both $k$ and the maximum degree $D$ of the trees are constant.

cs.DS↗

arXiv2002

Improved Phylogeny Comparisons: Non-Shared Edges Nearest Neighbor Interchanges, and Subtree Transfers

The number of the non-shared edges of two phylogenies is a basic measure of the dissimilarity between the phylogenies. The non-shared edges are also the building block for approximating a more sophisticated metric called the nearest neighbor interchange (NNI) distance. In this paper, we give the first subquadratic-time algorithm for finding the non-shared edges, which are then used to speed up the existing approximating algorithm for the NNI distance from $O(n^2)$ time to $O(n \log n)$ time. Another popular distance metric for phylogenies is the subtree transfer (STT) distance. Previous work on computing the STT distance considered degree-3 trees only. We give an approximation algorithm for the STT distance for degree-$d$ trees with arbitrary $d$ and with generalized STT operations.

cs.DS↗

arXiv2001

Predicting RNA Secondary Structures with Arbitrary Pseudoknots by Maximizing the Number of Stacking Pairs

The paper investigates the computational problem of predicting RNA secondary structures. The general belief is that allowing pseudoknots makes the problem hard. Existing polynomial-time algorithms are heuristic algorithms with no performance guarantee and can only handle limited types of pseudoknots. In this paper we initiate the study of predicting RNA secondary structures with a maximum number of stacking pairs while allowing arbitrary pseudoknots. We obtain two approximation algorithms with worst-case approximation ratios of 1/2 and 1/3 for planar and general secondary structures,respectively. For an RNA sequence of $n$ bases, the approximation algorithm for planar secondary structures runs in $O(n^3)$ time while that for the general case runs in linear time. Furthermore, we prove that allowing pseudoknots makes it NP-hard to maximize the number of stacking pairs in a planar secondary structure. This result is in contrast with the recent NP-hard results on psuedoknots which are based on optimizing some general and complicated energy functions.

cs.CE↗

arXiv2001

The Enhanced Double Digest Problem for DNA Physical Mapping

The double digest problem is a common NP-hard approach to constructing physical maps of DNA sequences. This paper presents a new approach called the enhanced double digest problem. Although this new problem is also NP-hard, it can be solved in linear time in certain theoretically interesting cases.

cs.CE↗

arXiv2001

Cavity Matchings, Label Compressions, and Unrooted Evolutionary Trees

We present an algorithm for computing a maximum agreement subtree of two unrooted evolutionary trees. It takes O(n^{1.5} log n) time for trees with unbounded degrees, matching the best known time complexity for the rooted case. Our algorithm allows the input trees to be mixed trees, i.e., trees that may contain directed and undirected edges at the same time. Our algorithm adopts a recursive strategy exploiting a technique called label compression. The backbone of this technique is an algorithm that computes the maximum weight matchings over many subgraphs of a bipartite graph as fast as it takes to compute a single matching.

cs.CE↗

arXiv2001

An Even Faster and More Unifying Algorithm for Comparing Trees via Unbalanced Bipartite Matchings

A widely used method for determining the similarity of two labeled trees is to compute a maximum agreement subtree of the two trees. Previous work on this similarity measure is only concerned with the comparison of labeled trees of two special kinds, namely, uniformly labeled trees (i.e., trees with all their nodes labeled by the same symbol) and evolutionary trees (i.e., leaf-labeled trees with distinct symbols for distinct leaves). This paper presents an algorithm for comparing trees that are labeled in an arbitrary manner. In addition to this generality, this algorithm is faster than the previous algorithms. Another contribution of this paper is on maximum weight bipartite matchings. We show how to speed up the best known matching algorithms when the input graphs are node-unbalanced or weight-unbalanced. Based on these enhancements, we obtain an efficient algorithm for a new matching problem called the hierarchical bipartite matching problem, which is at the core of our maximum agreement subtree algorithm.

cs.CV↗

arXiv2000

A Decomposition Theorem for Maximum Weight Bipartite Matchings

Let G be a bipartite graph with positive integer weights on the edges and without isolated nodes. Let n, N and W be the node count, the largest edge weight and the total weight of G. Let k(x,y) be log(x)/log(x^2/y). We present a new decomposition theorem for maximum weight bipartite matchings and use it to design an O(sqrt(n)W/k(n,W/N))-time algorithm for computing a maximum weight matching of G. This algorithm bridges a long-standing gap between the best known time complexity of computing a maximum weight matching and that of computing a maximum cardinality matching. Given G and a maximum weight matching of G, we can further compute the weight of a maximum weight matching of G-{u} for all nodes u in O(W) time.

cs.DS↗