Harikrishna Narasimhan

Predicting clinical response to anticancer drugs using an ex vivo platform that captures tumour heterogeneity

Predicting clinical response to anticancer drugs remains a major challenge in cancer treatment. Emerging reports indicate that the tumour microenvironment and heterogeneity can limit the predictive power of current biomarker-guided strategies for chemotherapy. Here we report the engineering of personalized tumour ecosystems that contextually conserve the tumour heterogeneity, and phenocopy the tumour microenvironment using tumour explants maintained in defined tumour grade-matched matrix support and autologous patient serum. The functional response of tumour ecosystems, engineered from 109 patients, to anticancer drugs, together with the corresponding clinical outcomes, is used to train a machine learning algorithm; the learned model is then applied to predict the clinical response in an independent validation group of 55 patients, where we achieve 100% sensitivity in predictions while keeping specificity in a desired high range. The tumour ecosystem and algorithm, together termed the CANScript technology, can emerge as a powerful platform for enabling personalized medicine.

Parallel artificial bee colony (PABC) algorithm

The artificial bee colony (ABC) algorithm is a metaheuristic algorithm for numerical optimization. It is based on the intelligent foraging behavior of honey bees. This paper presents a parallel version of the algorithm for shared memory architectures. The entire colony of bees is divided equally among the available processors. A set of solutions is placed in the local memory of each processor. A copy of each solution is also maintained in a global shared memory. During each cycle, the set of bees at a processor improves the solutions in the local memory. At the end of the cycle, the solutions are copied into the corresponding slots in the shared memory and made available to all other bees. It is shown that the proposed parallelization strategy does not degrade the quality of solutions obtained, but achieves substantial speedup.

SVM pAUC tight : a new support vector method for optimizing partial AUC based on a tight convex upper bound

The area under the ROC curve (AUC) is a well known performance measure in machine learning and data mining. In an increasing number of applications, however, ranging from ranking applications to a variety of important bioinformatics applications, performance is measured in terms of the partial area under the ROC curve between two specified false positive rates. In recent work, we proposed a structural SVM based approach for optimizing this performance measure (Narasimhan and Agarwal, 2013). In this paper, we develop a new support vector method, SVMpAUCtight, that optimizes a tighter convex upper bound on the partial AUC loss, which leads to both improved accuracy and reduced computational complexity. In particular, by rewriting the empirical partial AUC risk as a maximum over subsets of negative instances, we derive a new formulation, where a modified form of the earlier optimization objective is evaluated on each of these subsets, leading to a tighter hinge relaxation on the partial AUC loss. As with our previous method, the resulting optimization problem can be solved using a cutting-plane algorithm, but the new method has better run time guarantees. We also discuss a projected subgradient method for solving this problem, which offers additional computational savings in certain settings. We demonstrate on a wide variety of bioinformatics tasks, ranging from protein-protein interaction prediction to drug discovery tasks, that the proposed method does, in many cases, perform significantly better on the partial AUC measure than the previous structural SVM approach. In addition, we also develop extensions of our method to learn sparse and group sparse models, often of interest in biological applications.

Contribution-based clustering algorithm for content-based image retrieval

Clustering is a form of unsupervised classification that aims at grouping data points based on similarity. In this paper, we propose a new partitional clustering algorithm based on the notion of ‘contribution of a data point’. We apply the algorithm to content-based image retrieval and compare its performance with that of the k-means clustering algorithm. Unlike the k-means algorithm, our algorithm optimizes on both intra-cluster and inter-cluster similarity measures. It has three passes and each pass has the same time complexity as an iteration in the k-means algorithm. Our experiments on a bench mark image data set reveal that our algorithm improves on the recall at the cost of precision.

Game theoretic resistance to denial of service attacks using hidden difficulty puzzles

Denial of Service (DoS) vulnerabilities are one of the major concerns in today’s internet. Client-puzzles offer a good mechanism to defend servers against DoS attacks. In this paper, we introduce the notion of hidden puzzle difficulty, where the attacker cannot determine the difficulty of the puzzle without expending a minimal amount of computational resource. We propose three concrete puzzles that satisfy this requirement. Using game theory, we show that a defense mechanism is more effective when it uses a hidden difficulty puzzle. Based on the concept of Nash equilibrium, we develop suitable defense mechanisms that are better than the existing ones.

A randomized iterative improvement algorithm for photomosaic generation

A photomosaic is an image assembled from smaller images called tiles. When a photomosaic is viewed from a distance, it resembles a desired target image. The process of photomosaic generation can be viewed as an optimization problem, where a set of tiles needs to be arranged to resemble a target image. We impose a constraint on the number of times a tile image can be repeated in a photomosaic. A randomized iterative improvement algorithm is used to generate photomosaics and the intermediate results are used to produce interesting animations. We show that the proposed technique is more efficient than genetic algorithm.

Automatic summarization of cricket video events using genetic algorithm

Sports video mining has gained a lot of popularity in recent years. In this paper, we propose a flexible framework for summarization of cricket video content using genetic algorithm (GA). A summary of a cricket match should contain a variety of events, span the entire duration of the match and contain events of relatively high importance. These three parameters develop into a multi-objective optimization problem that is solved using genetic algorithm. Our experimental results on cricket matches are quite encouraging and we intend to extend this technique to other similar sports.

Cryptographic Approaches to Denial-of-Service Resistance

Authentication is a promising way to treat denial-of-service (DoS) threats against nonpublic services because it allows servers to restrict connections only to authorised users. However, there is a catch with this argument since authentication itself is typically a computationally intensive rocess that is necessarily exposed to unauthenticated entities. This means that the authentication protocol can become a source of denial-of-service vulnerability itself, thereby causing the same problem it is aimed at solving.

Support vector algorithms for optimizing the partial area under the roc curve

The area under the ROC curve (AUC) is a widely used performance measure in machine learning. Increasingly, however, in several applications, ranging from ranking to biometric screening to medicine, performance is measured not in terms of the full area under the ROC curve but in terms of the partial area under the ROC curve between two false-positive rates. In this letter, we develop support vector algorithms for directly optimizing the partial AUC between any two false-positive rates. Our methods are based on minimizing a suitable proxy or surrogate objective for the partial AUC error. In the case of the full AUC, one can readily construct and optimize convex surrogates by expressing the performance measure as a summation of pairwise terms. The partial AUC, on the other hand, does not admit such a simple decomposable structure, making it more challenging to design and optimize (tight) convex surrogates for this measure. Our approach builds on the structural SVM framework of Joachims (2005) to design convex surrogates for partial AUC and solves the resulting optimization problem using a cutting plane solver. Unlike the full AUC, where the combinatorial optimization needed in each iteration of the cutting plane solver can be decomposed and solved efficiently, the corresponding problem for the partial AUC is harder to decompose. One of our main contributions is a polynomial time algorithm for solving the combinatorial optimization problem associated with partial AUC. We also develop an approach for optimizing a tighter nonconvex hinge loss–based surrogate for the partial AUC using difference-of-convex programming. Our experiments on a variety of real-world and benchmark tasks confirm the efficacy of the proposed methods.

Moving average based predictors for MPEG-4 VBR traffic sources

Dynamic allocation of bandwidth allows us to maintain an agreed quality of service while streaming multimedia content in a network. This can be achieved by predicting future traffic in advance and allocating resources accordingly. In this paper, we propose new prediction techniques for MPEG-4 encoded variable bit rate video traffic based on the concept of moving average and extend them further using the gradient-descent approach. The resultant predictors are both simple and accurate and suitable for realtime prediction. We use the NLMS technique as a base-line predictor for our comparative study and show a performance improvement of about 11%.