arXiv subjects

Naveen Sivadasan

Publications and source records attributed to Naveen Sivadasan.

At least 19 recordsLinked to original sources

arXiv2026

Mitosis Detection in the Wild: Multi-Tumor and Context-Aware Generalization in the MIDOG 2025 Challenge

Automated mitosis detection is a well-established task in computational pathology. While previous benchmarks focused on scanner-induced domain shift, clinical "real-world" application requires models to be robust across the vast variance to be expected in the histological landscape. The MItosis DOmain Generalization (MIDOG) 2025 challenge was designed to evaluate algorithmic performance across unprecedented biological and contextual diversity. We curated a test dataset of 365 cases, encompassing 12 distinct human, canine and feline tumor types, digitized across multiple scanning platforms. Moving beyond hand-selected hotspots, the challenge required detection also in random tissue areas (representative of the whole slide detection situation) and challenging areas (areas rich in hard negatives). In the second track, we introduced the classification of atypical mitotic figures (AMFs). There were 18 teams submitting to the detection track, with F1 scores ranging up to 0.740. In the AMF detection track, we had 21 submissions with balanced accuracy values up to 0.908. Our analysis reveals that while most models perform reliably in traditional hotspots, significant performance degradation occurs in challenging ROIs, where false positive rates tripled. Furthermore, performance varied significantly across the 12 tumor types, highlighting "blind spots" in current state-of-the-art architectures when encountering rare or highly pleomorphic malignancies. Moreover, we evaluated the effectiveness of ensembling and found a mean increases of 1.5 and 1.3 percentage points in F1 score and balanced accuracy, respectively. In contrast, TTA showed no relevant improvement. MIDOG 2025 demonstrates that "in the wild" mitosis detection remains a significant hurdle. The transition from hotspot-only evaluation to a multi-contextual framework provides a more realistic proxy for clinical reliability.

cs.CV↗

arXiv2025

MIDOG 2025 Track 2: A Deep Learning Model for Classification of Atypical and Normal Mitotic Figures under Class and Hardness Imbalances

Motivation: Accurate classification of mitotic figures into normal and atypical types is crucial for tumor prognostication in digital pathology. However, developing robust deep learning models for this task is challenging due to the subtle morphological differences, as well as significant class and hardness imbalances in real-world histopathology datasets. Methods: We propose a novel deep learning approach based on a ResNet backbone with specialized classification heads. Our architecture uniquely models both the mitotic figure phenotype and the instance difficulty simultaneously. This method is specifically designed to handle the challenges of diverse tissue types, scanner variability, and imbalanced data. We employed focal loss to effectively mitigate the pronounced class imbalance, and a comprehensive data augmentation pipeline was implemented to enhance the model's robustness and generalizability. Results: Our approach demonstrated strong and consistent performance. In a 5-fold cross-validation on the MIDOG 2025 Track 2 dataset, it achieved a mean balanced accuracy of 0.8744 +/- 0.0093 and an ROC AUC of 0.9505 +/- 0.029. The model showed robust generalization across preliminary leaderboard evaluations, achieving an overall balanced accuracy of 0.8736 +/- 0.0204. Conclusion: The proposed method offers a reliable and generalizable solution for the classification of atypical and normal mitotic figures. By addressing the inherent challenges of real world data, our approach has the potential to support precise prognostic assessments in clinical practice and improve consistency in pathological diagnosis.

eess.IV↗

arXiv2022

Multi-label classification for biomedical literature: an overview of the BioCreative VII LitCovid Track for COVID-19 literature topic annotations

The COVID-19 pandemic has been severely impacting global society since December 2019. Massive research has been undertaken to understand the characteristics of the virus and design vaccines and drugs. The related findings have been reported in biomedical literature at a rate of about 10,000 articles on COVID-19 per month. Such rapid growth significantly challenges manual curation and interpretation. For instance, LitCovid is a literature database of COVID-19-related articles in PubMed, which has accumulated more than 200,000 articles with millions of accesses each month by users worldwide. One primary curation task is to assign up to eight topics (e.g., Diagnosis and Treatment) to the articles in LitCovid. Despite the continuing advances in biomedical text mining methods, few have been dedicated to topic annotations in COVID-19 literature. To close the gap, we organized the BioCreative LitCovid track to call for a community effort to tackle automated topic annotation for COVID-19 literature. The BioCreative LitCovid dataset, consisting of over 30,000 articles with manually reviewed topics, was created for training and testing. It is one of the largest multilabel classification datasets in biomedical scientific literature. 19 teams worldwide participated and made 80 submissions in total. Most teams used hybrid systems based on transformers. The highest performing submissions achieved 0.8875, 0.9181, and 0.9394 for macro F1-score, micro F1-score, and instance-based F1-score, respectively. The level of participation and results demonstrate a successful track and help close the gap between dataset curation and method development. The dataset is publicly available via https://ftp.ncbi.nlm.nih.gov/pub/lu/LitCovid/biocreative/ for benchmarking and further development.

cs.DL↗

arXiv2021

Adiabatic Quantum Feature Selection for Sparse Linear Regression

Linear regression is a popular machine learning approach to learn and predict real valued outputs or dependent variables from independent variables or features. In many real world problems, its beneficial to perform sparse linear regression to identify important features helpful in predicting the dependent variable. It not only helps in getting interpretable results but also avoids overfitting when the number of features is large, and the amount of data is small. The most natural way to achieve this is by using `best subset selection' which penalizes non-zero model parameters by adding $\ell_0$ norm over parameters to the least squares loss. However, this makes the objective function non-convex and intractable even for a small number of features. This paper aims to address the intractability of sparse linear regression with $\ell_0$ norm using adiabatic quantum computing, a quantum computing paradigm that is particularly useful for solving optimization problems faster. We formulate the $\ell_0$ optimization problem as a Quadratic Unconstrained Binary Optimization (QUBO) problem and solve it using the D-Wave adiabatic quantum computer. We study and compare the quality of QUBO solution on synthetic and real world datasets. The results demonstrate the effectiveness of the proposed adiabatic quantum computing approach in finding the optimal solution. The QUBO solution matches the optimal solution for a wide range of sparsity penalty values across the datasets.

cs.LG↗

arXiv2016

Kmerlight: fast and accurate k-mer abundance estimation

k-mers (nucleotide strings of length k) form the basis of several algorithms in computational genomics. In particular, k-mer abundance information in sequence data is useful in read error correction, parameter estimation for genome assembly, digital normalization etc. We give a streaming algorithm Kmerlight for computing the k-mer abundance histogram from sequence data. Our algorithm is fast and uses very small memory footprint. We provide analytical bounds on the error guarantees of our algorithm. Kmerlight can efficiently process genome scale and metagenome scale data using standard desktop machines. Few applications of abundance histograms computed by Kmerlight are also shown. We use abundance histogram for de novo estimation of repetitiveness in the genome based on a simple probabilistic model that we propose. We also show estimation of k-mer error rate in the sampling using abundance histogram. Our algorithm can also be used for abundance estimation in a general streaming setting. The Kmerlight tool is written in C++ and is available for download and use from https://github.com/nsivad/kmerlight.

cs.DS↗

arXiv2015

A cubic vertex kernel for Diamond-free Edge Deletion and more

A diamond is a graph obtained by removing an edge from a complete graph on four vertices. A graph is diamond-free if it does not contain an induced diamond. The Diamond-free Edge Deletion problem asks whether there exist at most $k$ edges in the input graph $G$ whose deletion results in a diamond-free graph. For this problem, a polynomial kernel of $O(k^4$) vertices was found by Fellows et. al. (Discrete Optimization, 2011). In this paper, we give an improved kernel of $O(k^3)$ vertices for Diamond-free Edge Deletion. Further, we give an $O(k^2)$ vertex kernel for a related problem {Diamond,K_t}-free Edge Deletion, where $t\geq 4$ is any fixed integer. To complement our results, we prove that these problems are NP-complete even for $K_4$-free graphs and can be solved neither in subexponential time (i.e., $2^{o(|G|)}$) nor in parameterized subexponential time (i.e., $2^{o(k)}\cdot |G|^{O(1)}$), unless Exponential Time Hypothesis fails. Our reduction implies the hardness and lower bound for a general class of problems, where these problems come as a special case.

cs.DS↗

arXiv2015

Subgraph Similarity Search in Large Graphs

One of the major challenges in applications related to social networks, computational biology, collaboration networks etc., is to efficiently search for similar patterns in their underlying graphs. These graphs are typically noisy and contain thousands of vertices and millions of edges. In many cases, the graphs are unlabeled and the notion of similarity is also not well defined. We study the problem of searching an induced subgraph in a large target graph that is most similar to the given query graph. We assume that the query graph and target graph are undirected and unlabeled. We use graphlet kernels \cite{shervashidze2009efficient} to define graph similarity. Graphlet kernels are known to perform better than other kernels in different applications. Our algorithm maps topological neighborhood information of vertices in the query and target graphs to vectors. These local topological informations are then combined to find a target subgraph having highly similar global topology with the given query graph. We tested our algorithm on several real world networks such as facebook network, google plus network, youtube network, amazon network etc. Most of them contain thousands of vertices and million edges. Our algorithm is able to detect highly similar matches when queried in these networks. Our multi-threaded implementation takes about one second to find the match on a 32 core machine, excluding the time for one time preprocessing. Computationally expensive parts of our algorithm can be further scaled to standard parallel and distributed frameworks like map-reduce.

cs.SI↗

arXiv2015

Parameterized Lower Bounds and Dichotomy Results for the NP-completeness of $H$-free Edge Modification Problems

For a graph $H$, the $H$-free Edge Deletion problem asks whether there exist at most $k$ edges whose deletion from the input graph $G$ results in a graph without any induced copy of $H$. $H$-free Edge Completion and $H$-free Edge Editing are defined similarly where only completion (addition) of edges are allowed in the former and both completion and deletion are allowed in the latter. We completely settle the classical complexities of these problems by proving that $H$-free Edge Deletion is NP-complete if and only if $H$ is a graph with at least two edges, $H$-free Edge Completion is NP-complete if and only if $H$ is a graph with at least two non-edges and $H$-free Edge Editing is NP-complete if and only if $H$ is a graph with at least three vertices. Additionally, we prove that, these NP-complete problems cannot be solved in parameterized subexponential time, i.e., in time $2^{o(k)}\cdot |G|^{O(1)}$, unless Exponential Time Hypothesis fails. Furthermore, we obtain implications on the incompressibility of these problems.

cs.DS↗

arXiv2015

Parameterized lower bound and NP-completeness of some $H$-free Edge Deletion problems

For a graph $H$, the $H$-free Edge Deletion problem asks whether there exist at most $k$ edges whose deletion from the input graph $G$ results in a graph without any induced copy of $H$. We prove that $H$-free Edge Deletion is NP-complete if $H$ is a graph with at least two edges and $H$ has a component with maximum number of vertices which is a tree or a regular graph. Furthermore, we obtain that these NP-complete problems cannot be solved in parameterized subexponential time, i.e., in time $2^{o(k)}\cdot |G|^{O(1)}$, unless Exponential Time Hypothesis fails.

cs.DS↗

arXiv2014

The chromatic discrepancy of graphs

For a proper vertex coloring $c$ of a graph $G$, let $φ_c(G)$ denote the maximum, over all induced subgraphs $H$ of $G$, the difference between the chromatic number $χ(H)$ and the number of colors used by $c$ to color $H$. We define the chromatic discrepancy of a graph $G$, denoted by $φ(G)$, to be the minimum $φ_c(G)$, over all proper colorings $c$ of $G$. If $H$ is restricted to only connected induced subgraphs, we denote the corresponding parameter by $\hatφ(G)$. These parameters are aimed at studying graph colorings that use as few colors as possible in a graph and all its induced subgraphs. We study the parameters $φ(G)$ and $\hatφ(G)$ and obtain bounds on them. We obtain general bounds, as well as bounds for certain special classes of graphs including random graphs. We provide structural characterizations of graphs with $φ(G) = 0$ and graphs with $\hatφ(G) = 0$. We also show that computing these parameters is NP-hard.

math.CO↗

arXiv2014

On Polynomial Kernelization of $\mathcal{H}$-free Edge Deletion

For a set of graphs $\mathcal{H}$, the \textsc{$\mathcal{H}$-free Edge Deletion} problem asks to find whether there exist at most $k$ edges in the input graph whose deletion results in a graph without any induced copy of $H\in\mathcal{H}$. In \cite{cai1996fixed}, it is shown that the problem is fixed-parameter tractable if $\mathcal{H}$ is of finite cardinality. However, it is proved in \cite{cai2013incompressibility} that if $\mathcal{H}$ is a singleton set containing $H$, for a large class of $H$, there exists no polynomial kernel unless $coNP\subseteq NP/poly$. In this paper, we present a polynomial kernel for this problem for any fixed finite set $\mathcal{H}$ of connected graphs and when the input graphs are of bounded degree. We note that there are \textsc{$\mathcal{H}$-free Edge Deletion} problems which remain NP-complete even for the bounded degree input graphs, for example \textsc{Triangle-free Edge Deletion}\cite{brugmann2009generating} and \textsc{Custer Edge Deletion($P_3$-free Edge Deletion)}\cite{komusiewicz2011alternative}. When $\mathcal{H}$ contains $K_{1,s}$, we obtain a stronger result - a polynomial kernel for $K_t$-free input graphs (for any fixed $t> 2$). We note that for $s>9$, there is an incompressibility result for \textsc{$K_{1,s}$-free Edge Deletion} for general graphs \cite{cai2012polynomial}. Our result provides first polynomial kernels for \textsc{Claw-free Edge Deletion} and \textsc{Line Edge Deletion} for $K_t$-free input graphs which are NP-complete even for $K_4$-free graphs\cite{yannakakis1981edge} and were raised as open problems in \cite{cai2013incompressibility,open2013worker}.

cs.DS↗

arXiv2014

Approximating the Cubicity of Trees

Cubicity of a graph $G$ is the smallest dimension $d$, for which $G$ is a unit disc graph in ${\mathbb{R}}^d$, under the $l^\infty$ metric, i.e. $G$ can be represented as an intersection graph of $d$-dimensional (axis-parallel) unit hypercubes. We call such an intersection representation a $d$-dimensional cube representation of $G$. Computing cubicity is known to be inapproximable in polynomial time, within an $O(n^{1-ε})$ factor for any $ε>0$, unless NP=ZPP. In this paper, we present a randomized algorithm that runs in polynomial time and computes cube representations of trees, of dimension within a constant factor of the optimum. It is also shown that the cubicity of trees can be approximated within a constant factor in deterministic polynomial time, if the cube representation is not required to be computed. As far as we know, this is the first constant factor approximation algorithm for computing the cubicity of trees. It is not yet clear whether computing the cubicity of trees is NP-hard or not.

cs.DM↗

arXiv2012

Lower Bounds for Boxicity

An axis-parallel $b$-dimensional box is a Cartesian product $R_1\times R_2\times...\times R_b$ where $R_i$ is a closed interval of the form $[a_i,b_i]$ on the real line. For a graph $G$, its \emph{boxicity} box(G) is the minimum dimension $b$, such that $G$ is representable as the intersection graph of boxes in $b$-dimensional space. Although boxicity was introduced in 1969 and studied extensively, there are no significant results on lower bounds for boxicity. In this paper, we develop two general methods for deriving lower bounds. Applying these methods we give several results, some of which are listed below: (1) The boxicity of a graph on $n$ vertices with no universal vertices and minimum degree $δ$ is at least $n/2(n-δ-1)$. (2) Consider the $\mathcal{G}(n,p)$ model of random graphs. Let $ p \le 1- \frac{40 \log n}{n^2}$. Then, for $G \in \mathcal{G}(n,p)$, almost surely $box(G)=Ω(np(1-p))$. On setting $p=1/2$ we immediately infer that almost all graphs have boxicity $Ω(n)$. (3) Spectral lower bounds for the boxicity of $k$-regular graphs. (4) The boxicity of random $k$-regular graphs on $n$ vertices (where $k$ is fixed) is $Ω(k/\log k)$. (5) There exists a positive constant$c$ such that almost all balanced bipartite graphs on $2n$ vertices with exactly $m$ edges have boxicity at least $c m/n$, for $ m\le c' n^2/3$ for any positive constant $c' < 1$.

math.CO↗

arXiv2010

Boxicity of Line Graphs

Boxicity of a graph H, denoted by box(H), is the minimum integer k such that H is an intersection graph of axis-parallel k-dimensional boxes in R^k. In this paper, we show that for a line graph G of a multigraph, box(G) <= 2Δ(\lceil log_2(log_2(Δ)) \rceil + 3) + 1, where Δdenotes the maximum degree of G. Since Δ<= 2(χ- 1), for any line graph G with chromatic number χ, box(G) = O(χlog_2(log_2(χ))). For the d-dimensional hypercube H_d, we prove that box(H_d) >= (\lceil log_2(log_2(d)) \rceil + 1)/2. The question of finding a non-trivial lower bound for box(H_d) was left open by Chandran and Sivadasan in [L. Sunil Chandran and Naveen Sivadasan. The cubicity of Hypercube Graphs. Discrete Mathematics, 308(23):5795-5800, 2008]. The above results are consequences of bounds that we obtain for the boxicity of fully subdivided graphs (a graph which can be obtained by subdividing every edge of a graph exactly once).

math.CO↗

arXiv2008

On the cubicity of bipartite graphs

{\it A unit cube in $k$-dimension (or a $k$-cube) is defined as the cartesian product $R_1 \times R_2 \times ... \times R_k$, where each $R_i$ is a closed interval on the real line of the form $[a_i, a_i+1]$. The {\it cubicity} of $G$, denoted as $cub(G)$, is the minimum $k$ such that $G$ is the intersection graph of a collection of $k$-cubes. Many NP-complete graph problems can be solved efficiently or have good approximation ratios in graphs of low cubicity. In most of these cases the first step is to get a low dimensional cube representation of the given graph. It is known that for a graph $G$, $cub(G) \leq \lfloor\frac{2n}{3}\rfloor$. Recently it has been shown that for a graph $G$, $cub(G) \leq 4(Δ+ 1)\ln n$, where $n$ and $Δ$ are the number of vertices and maximum degree of $G$, respectively. In this paper, we show that for a bipartite graph $G = (A \cup B, E)$ with $|A| = n_1$, $|B| = n_2$, $n_1 \leq n_2$, and $Δ' = \min\{Δ_A, Δ_B\}$, where $Δ_A = {max}_{a \in A}d(a)$ and $Δ_B = {max}_{b \in B}d(b)$, $d(a)$ and $d(b)$ being the degree of $a$ and $b$ in $G$ respectively, $cub(G) \leq 2(Δ'+2) \lceil \ln n_2 \rceil$. We also give an efficient randomized algorithm to construct the cube representation of $G$ in $3(Δ'+2)\lceil \ln n_2 \rceil$ dimensions. The reader may note that in general $Δ'$ can be much smaller than $Δ$.}

cs.DM↗

arXiv2008

JDATATRANS for Array Obfuscation in Java Source Code to Defeat Reverse Engineering from Decompiled Codes

Software obfuscation or obscuring a software is an approach to defeat the practice of reverse engineering a software for using its functionality illegally in the development of another software. Java applications are more amenable to reverse engineering and re-engineering attacks through methods such as decompilation because Java class files store the program in a semi complied form called 'byte' codes. The existing obfuscation systems obfuscate the Java class files. Obfuscated source code produce obfuscated byte codes and hence two level obfuscation (source code and byte code level) of the program makes it more resilient to reverse engineering attacks. But source code obfuscation is much more difficult due to richer set of programming constructs and the scope of the different variables used in the program and only very little progress has been made on this front. Hence programmers resort to adhoc manual ways of obscuring their program which makes it difficult for its maintenance and usability. To address this issue partially, we developed a user friendly tool JDATATRANS to obfuscate Java source code by obscuring the array usages. Using various array restructuring techniques such as 'array splitting', 'array folding' and 'array flattening', in addition to constant hiding, our system obfuscate the input Java source code and produce an obfuscated Java source code that is functionally equivalent to the input program. We also perform a number of experiments to measure the potency, resilience and cost incurred by our tool.

cs.CR↗

arXiv2008

On the cubicity of AT-free graphs and circular-arc graphs

A unit cube in $k$ dimensions ($k$-cube) is defined as the the Cartesian product $R_1\times R_2\times...\times R_k$ where $R_i$(for $1\leq i\leq k$) is a closed interval of the form $[a_i,a_i+1]$ on the real line. A graph $G$ on $n$ nodes is said to be representable as the intersection of $k$-cubes (cube representation in $k$ dimensions) if each vertex of $G$ can be mapped to a $k$-cube such that two vertices are adjacent in $G$ if and only if their corresponding $k$-cubes have a non-empty intersection. The \emph{cubicity} of $G$ denoted as $\cubi(G)$ is the minimum $k$ for which $G$ can be represented as the intersection of $k$-cubes. We give an $O(bw\cdot n)$ algorithm to compute the cube representation of a general graph $G$ in $bw+1$ dimensions given a bandwidth ordering of the vertices of $G$, where $bw$ is the \emph{bandwidth} of $G$. As a consequence, we get $O(Δ)$ upper bounds on the cubicity of many well-known graph classes such as AT-free graphs, circular-arc graphs and co-comparability graphs which have $O(Δ)$ bandwidth. Thus we have: 1) $\cubi(G)\leq 3Δ-1$, if $G$ is an AT-free graph. 2) $\cubi(G)\leq 2Δ+1$, if $G$ is a circular-arc graph. 3) $\cubi(G)\leq 2Δ$, if $G$ is a co-comparability graph. Also for these graph classes, there are constant factor approximation algorithms for bandwidth computation that generate orderings of vertices with $O(Δ)$ width. We can thus generate the cube representation of such graphs in $O(Δ)$ dimensions in polynomial time.

cs.DM↗

arXiv2008

Representing graphs as the intersection of axis-parallel cubes

A unit cube in $k$ dimensional space (or \emph{$k$-cube} in short) is defined as the Cartesian product $R_1\times R_2\times...\times R_k$ where $R_i$(for $1\leq i\leq k$) is a closed interval of the form $[a_i,a_i+1]$ on the real line. A $k$-cube representation of a graph $G$ is a mapping of the vertices of $G$ to $k$-cubes such that two vertices in $G$ are adjacent if and only if their corresponding $k$-cubes have a non-empty intersection. The \emph{cubicity} of $G$, denoted as $\cubi(G)$, is the minimum $k$ such that $G$ has a $k$-cube representation. Roberts \cite{Roberts} showed that for any graph $G$ on $n$ vertices, $\cubi(G)\leq 2n/3$. Many NP-complete graph problems have polynomial time deterministic algorithms or have good approximation ratios in graphs of low cubicity. In most of these algorithms, computing a low dimensional cube representation of the given graph is usually the first step. We present an efficient algorithm to compute the $k$-cube representation of $G$ with maximum degree $Δ$ in $O(Δ\ln b)$ dimensions where $b$ is the bandwidth of $G$. Bandwidth of $G$ is at most $n$ and can be much lower. The algorithm takes as input a bandwidth ordering of the vertices in $G$. Though computing the bandwidth ordering of vertices for a graph is NP-hard, there are heuristics that perform very well in practice. Even theoretically, there is an $O(\log^4 n)$ approximation algorithm for computing the bandwidth ordering of a graph using which our algorithm can produce a $k$-cube representation of any given graph in $k=O(Δ(\ln b + \ln\ln n))$ dimensions. Both the bounds on cubicity are shown to be tight upto a factor of $O(\log\log n)$.

cs.DM↗