We address the problem of serving Deep Neural Networks (DNNs) efficiently from a cluster GPUs. In order to realize promise very low-cost processing made by accelerators such as GPUs, it is essential run them at sustained high utilization. Doing so requires cluster-scale resource management that performs detailed scheduling reasoning about groups DNN invocations need be co-scheduled, and moving conventional whole-DNN execution model executing fragments DNNs. Nexus fully implemented system...

Many emerging scientific and industrial applications require transferring multiple Tbytes of data on a daily basis. Examples include pushing from particle accelerators/colliders to laboratories around the world, synchronizing data-centers across continents, replicating collections high definition videos events taking place at different time-zones. A key property all above is their ability tolerate delivery delays ranging few hours days. Such Delay Tolerant Bulk (DTB) are currently being...

In this paper, we address the problem of cross-layer denial service attack in wireless data networks. We introduce SPREAD -a novel adaptive diversification approach to provide resiliency against such attacks. relies on a mechanism-hopping technique, which can be seen as multi-layer extension frequency-hopping technique. apply game-theoretic framework for modeling interaction communicating nodes and adversaries analyze proposed approach. reason about advantages various types jammers...

Recently, cloud computing services such as Amazon EC2 have used virtualization to provide customers with virtual machines running on the provider's hardware, typically charging by wall clock time rather than resources consumed. Under this business model, manipulation of scheduler may allow theft-of-service at expense other customers. We discovered and implemented an attack scenario which when allowed consume more CPU regardless fair share. a novel analysis necessary conditions for attacks,...

Many emerging scientific and industrial applications require transferring multiple terabytes of data on a daily basis. Examples include pushing from particle accelerators/colliders to laboratories around the world, synchronizing datacenters across continents, replicating collections high-definition videos events taking place at different time-zones. A key property all above is their ability tolerate delivery delays ranging few hours days. Such delay-tolerant bulk (DTB) are currently being...

There have been significant advances in our understanding of cancer as a disease at the molecular level. Combined with improved diagnostic systems, concept personalized medicine was introduced where therapy for every patient can be customized according to their profile. The nanotechnology approach formulation design and advent drug delivery systems small molecules biologics has contributed development medicine. Despite progress, effective management treatment remains clinical challenge....

Consider public health officials aiming to spread awareness about a new vaccine in community interconnected by social network. How can they distribute information with minimal resources, so as avoid polarization and ensure community-wide convergence of opinion? To tackle such challenges, we initiate the study sample complexity opinion formation networks. Our framework is built on recognized game, where regard each agent’s data-derived model, unlike previous works that treat opinions...

We introduce a notion of universality in the context optimization problems with partial information. Universality is framework for dealing uncertainty by guaranteeing certain quality goodness all possible completions information set. Universal variants can be defined that are both natural and well-motivated. consider universal versions three classical problems: TSP, Steiner Tree Set Cover.We present polynomial-time algorithm to find tour on given metric space over n vertices such any subset...

A graph $G$ is said to be $d$-distinguishable if there a $d$-coloring of which no non-trivial automorphism preserves. That is, $\exists \chi: G \rightarrow \{1, \ldots, d\},$ $$ \forall \phi \in \mathrm{Aut}(G) \setminus \{\mathbf{id}\}, \exists v, \chi(v) \neq \chi(\phi(v)). It was conjectured that $|G| > |\mathrm{Aut}(G)|$ and the $\mathrm{Aut}(G)$ action on has singleton orbits, then 2-distinguishable. We give an example where this fails. partially repair conjecture by showing when...

Motivated by applications in social networks, peer-to-peer and overlay we define study the Bounded Budget Connection (BBC) game - have a collection of n players or nodes each whom has budget for purchasing links; link cost as well length node set preference weights remaining nodes; objective is to use its buy outgoing links so minimize sum preference-weighted distances nodes. We structural complexity-theoretic properties pure Nash equilibria BBC games. show that determining existence...

Blind rendezvous is a fundamental problem in cognitive radio networks. The involves collection of agents (radios) that wish to discover each other (i.e., rendezvous) the blind setting where there no shared infrastructure and they initially have knowledge other. Time divided into discrete slots spectrum channels, [n] = 1, 2, ..., n. Each agent may access (or hop on) single channel time slot two when on same slot. goal design deterministic hopping schedules for so as guarantee between any pair...

We study the problem of finding small trees. Classical network design problems are considered with additional constraint that only a specified number $k$ nodes required to be connected in solution. A prototypical example is $k$MST which we require tree minimum weight spanning at least an edge-weighted graph. show NP-hard even for points Euclidean plane. provide approximation algorithms performance ratio $2\sqrt{k}$ general case and $O(k^{1/4})$ Polynomial-time exact solutions also presented...

In hardware virtualization a hypervisor provides multiple Virtual Machines (VMs) on single physical system, each executing separate operating system instance. The schedules execution of these VMs much as the scheduler in an does, balancing factors such fairness d I/O performance. As may be vulnerable to malicious behavior part users seeking deny service others or maximize their own resource usage. Recently, publically available cloud computing services Amazon EC2 have used provide customers...

We resolve the computational complexity of a number outstanding open problems with practical applications. Here is list we show to be ${\bf{PPAD}}$-complete, along domains significance: fractional stable paths problem (FSPP)---Internet routing; core balanced games---economics and game theory; Scarf's lemma---combinatorics; hypergraph matching---social choice preference systems; bounded budget connection games (FBBC)---social networks; strong kernel---graph theory. In fact, that no fully...

Credit card fraud transaction detection system is a method used for determining the fraudulent transactions that take place every once in while. The project uses test data set of around 27,000 credit which have been taken from Caltech (Kaggle). comprises primarily 2 major algorithms and anomaly as to classify transactions. Local Outlier Factor (LoF) defines various parameters be criteria It then checks upon different present given LoF. This factor gives each score based on or will place....

A flow of a commodity is said to be confluent if at any node all the leaves along single edge. In this article, we study single-commodity problems, where need route given demands destination using flow. Single- and multi-commodity flows arise in variety application areas, most notably networking; fact, Internet are (multi-commodity) since routing based. We present near-tight approximation algorithms, hardness results, existence theorems for minimizing congestion flows. The maximum edge...

In a landmark paper, Papadimitriou introduced number of syntactic subclasses TFNP based on proof styles that (unlike TFNP) admit complete problems. A recent series results has shown finding Nash equilibria is for PPAD, particularly notable subclass TFNP. major goal this work to expand the universe known PPAD-complete We resolve computational complexity outstanding open problems with practical applications. Here list we show be PPAD-complete, along domains significance: Fractional Stable...

In hardware virtualization a hypervisor provides multiple Virtual Machines (VMs) on single physical system, each executing separate operating system instance. The schedules execution of these VMs much as the scheduler in an does, balancing factors such fairness and I/O performance. As may be vulnerable to malicious behavior part users seeking deny service others or maximize their own resource usage. Recently, publically available cloud computing services Amazon EC2 have used provide...

Real-world networks often consist of multiple layers, be they infrastructure such as airline or social collaboration networks. A common aspect to these is that there are sub-networks evolve in parallel on the same node set -- referred multiplex For example, case networks, cities (nodes) have been well-established for several decades if not centuries, but over time new airlines (sub-networks) emerge and each creates its own flight linkages between cities. Similarly modalities communications...