Research Initiatives Projects

Limiting Manipulation in Data Centers and the Cloud

Researchers are designing algorithms to allocate resources in datacenters and clouds that can't be manipulated by users. In datacenters and clouds, computing resources or individual machines are allocated to users based on the requirements of the jobs they want to run. Users can manipulate allocations by misreporting their requirements. In this project, researchers design algorithms that are less susceptible to such manipulation. They will also use algorithmic mechanism design and game theory to develop general procedures for converting protocols so that they can't be manipulated.

Evaluating Price Mechanisms for Clouds

Researchers are studying the problems that arise in cloud computing centers that use economic models to allocate resources. In these clouds, resources, such as storage, processing, and data transfer, must be allocated to different users. In economics-based clouds, artificial economies are set up; each resource is assigned a "price" and each user is given a "budget," which they spend on the resources they need.

Maven (Malleable Array of Vector-thread ENgines)

In earlier work at MIT, Professor Asanovic's team developed the Scale vector-thread architecture and processor prototype, which combines data-level and thread-level parallel execution models in a single unified architecture. MAVEN is the second-generation vector-thread architecture, designed to scale up to hundreds of execution lanes, and with the goal of providing very high throughput at low energy for a wide variety of parallel applications.

Monolithic Silicon Photonics for Processor-to-DRAM Interconnects

In a collaboration with the MIT Center for Integrated Photonic Systems, researchers from the Architecture Group are exploring the use of silicon photonics for processor-to-memory interconnect. Projected advances in electrical signaling seem unlikely to fulfill the memory bandwidth demands of future manycore processor chips. Monolithic silicon photonics, which integrates optical components with electrical transistors in a conventional CMOS process, is a promising new technology that could provide large improvements in achievable interconnect bandwidth.

Stochastic Direct Reinforcment Algorithms

Researchers at ICSI are developing Stochastic Direct Reinforcement algorithms, which show promise of being a superior alternative to traditional reinforcement learning methods for solving real world applications.

Analysis of Heuristic Combinatorial Algorithms

In many practical situations heuristic algorithms reliably give satisfactory solutions to real-life instances of optimization problems, despite evidence from computational complexity theory that the problems are intractable. Our goal is to understand this seeming contradiction, and to put the construction and evaluation of heuristic algorithms on a firmer footing. We will develop a general empirical method for selecting an optimal choice of parameters and subroutines within a well defined heuristic algorithmic strategy.

Finding Conserved Protein Modules

A long-term goal of computational molecular biology is to extract, from large data sets, information about how proteins work together to carry out life processes at a cellular level. We are investigating protein-protein interaction (PPI) networks, in which the vertices are the proteins within a species and the edges indicate direct interactions between proteins. Our goal is to discover conserved protein modules: richly interacting sets of proteins whose patterns of interaction are conserved across two or more species.

Transcriptional Regulation

Dissection of regulatory networks that conrol gene transcription is one of the greatest challenges of functional genomics. The Algorithms Group addressed the problem of modeling generic features of structurally but not textually related DNA motifs. The work divides into several parts: (1) A new approach to the recognition of transcription-factor binding sites, based on the principle that transcription factors divide naturally into families, and that the binding site motifs for transcription factors within the same family have common features.

Methods for the Analysis of High-Throughput Sequencing Data

We are currently developing methods for the design and analysis of studies that involve high-throughput sequencing technologies, such as the Solexa, 454, or Solid platforms.

Statistical Genetics and Populaton Genetics

We develop computational methods for the inference of evolutionary and genetic characteristics, such as the inference of recombination events, estimation of mutation rates, human history, etc., from genetic data.

Computational Methods for the Identification of Disease-Genotype Associations

We develop computational methods that aid in the analysis of genome-wide association studies, or other studies that involve the inference of a relation between a genetic variant such as an SNP or a copy-number variant (CNV), with a given phenotype that was measured for the studied population. These methods include haplotype inference methods, ancestry inference methods, and the incorporation of these in a statistical or machine-learning framework that is used to test for an association of a genetic marker with a phenotype.

Analysis of Genome-Wide Association Studies for Common Diseases

In these studies, sets of cases (individuals carrying a disease) and controls (background population) are collected and genotyped for genetic variants, normally single nucleotide polymorphisms (SNPs). Our group is collaborating closely with groups of geneticists and epidimiologists who have collected such samples. We take part in the analysis of these studies, and in some cases also in the design of the studies.