https://www2.econ.iastate.edu/tesfatsi/DeepLearningInNeuralNetworksOverview.JSchmidhuber2015.pdf

Deep learning in neural networks

https://www.cs.jhu.edu/~misha/ReadingSeminar/Papers/Bay08.pdf

Speeded-Up Robust Features (SURF)

We will review some of the major results in random graphs and some of the more challenging open problems. We will cover algorithmic and structural questions. We will touch on newer models, including those related to the WWW.

Random graphs

/pdf/convex-analysisnoer-san-nojin-zhan-nituite-5dl2rbmoyr.pdf

Convex Analysisの二,三の進展について

Models for the processes by which ideas and influence propagate through a social network have been studied in a number of domains, including the diffusion of medical and technological innovations, the sudden and widespread adoption of various strategies in game-theoretic settings, and the effects of "word of mouth" in the promotion of new products. Recently, motivated by the design of viral marketing strategies, Domingos and Richardson posed a fundamental algorithmic problem for such social network processes: if we can try to convince a subset of individuals to adopt a new product or innovation, and the goal is to trigger a large cascade of further adoptions, which set of individuals should we target?We consider this problem in several of the most widely studied models in social network analysis. The optimization problem of selecting the most influential nodes is NP-hard here, and we provide the first provable approximation guarantees for efficient algorithms. Using an analysis framework based on submodular functions, we show that a natural greedy strategy obtains a solution that is provably within 63% of optimal for several classes of models; our framework suggests a general approach for reasoning about the performance guarantees of algorithms for these types of influence problems in social networks.We also provide computational experiments on large collaboration networks, showing that in addition to their provable guarantees, our approximation algorithms significantly out-perform node-selection heuristics based on the well-studied notions of degree centrality and distance centrality from the field of social networks.

/pdf/maximizing-the-spread-of-influence-through-a-social-network-47y1ehuvwi.pdf

Maximizing the Spread of Influence through a Social Network

The $k$-core decomposition is a fundamental primitive in many machine learning and data mining applications. We present the first distributed and the first streaming algorithms to compute and maintain an approximate $k$-core decomposition with provable guarantees. Our algorithms achieve rigorous bounds on space complexity while bounding the number of passes or number of rounds of computation. We do so by presenting a new powerful sketching technique for $k$-core decomposition, and then by showing it can be computed efficiently in both streaming and MapReduce models. Finally, we confirm the effectiveness of our sketching technique empirically on a number of publicly available graphs.

/pdf/parallel-and-streaming-algorithms-for-k-core-decomposition-k77wgf7qxp.pdf

Parallel and Streaming Algorithms for K-Core Decomposition

Stars: Tera-Scale Graph Building for Clustering and Learning

Internet advertisers (buyers) repeatedly procure ad impressions from ad platforms (sellers) with the aim to maximize total conversion (i.e. ad value) while respecting both budget and return-on-investment (ROI) constraints for efficient utiliza-tion of limited monetary resources. Facing such a constrained buyer who aims to learn her optimal strategy to acquire impressions, we study from a seller’s perspective how to learn and price ad impressions through repeated posted price mechanisms to maximize revenue. For this two-sided learning setup, we propose a learning algorithm for the seller that utilizes an episodic binary-search procedure to identify a revenue-optimal selling price. We show that such a simple learning algorithm enjoys low seller regret when within each episode, the budget and ROI constrained buyer approximately best responds to the posted price. We present simple yet natural buyer’s bidding algorithms under which the buyer approximately best responds while satisfying budget and ROI constraints, leading to a low regret for our proposed seller pricing algorithm. The design of our seller algorithm is motivated by the fact that the seller’s revenue function admits a bell-shaped structure when the buyer best responds to prices under budget and ROI constraints, enabling our seller algorithm to identify revenue-optimal selling prices efficiently.

Pricing against a Budget and ROI Constrained Buyer

Due to the increasing security threats in the Internet, new overlay network architectures have been proposed to secure privileged services. In these architectures, the application servers are protected by a defense perimeter where only traffic from entities called servelets are allowed to pass. End users must be authorized and can only communicate with entities called access points (APs). APs relay authorized users' requests to servelets, which in turn pass them to the servers. The identity of APs are publicly known while the servelets are typically secret. All communications are done through the public Internet. Thus all the entities involved forms an overlay network. The main component of this distributed system consists of n APs. and m servelets. A design for a network is a bipartite graph with APs on one side, and the servelets on the other side. If an AP is compromised by an attacker, all the servelets that are connected to it are subject to attack. An AP is blocked, if all servelets connected to it are subject to attack. We consider two models for the failures: In the average case model, we assume that each AP i fails with a given probability pi. In the worst case model, we assume that there is an adversary that knowing the topology of the network, chooses at most k APs to compromise. In both models, our objective is to design the connections between APs and servelets to minimize the (expected/worst-case) number of blocked APs. In this paper, we give a polynomial-time algorithm for this problem in the average-case model when the number of servelets is a constant. We also show that if the probability of failure of each AP is at least 1/2, then in the optimal design each AP is connected to only one servelet (we call such designs star-shaped), and give a polynomial-time algorithm to find the best star-shaped design. We observe that this statement is not true if the failure probabilities are small. In the worst-case model, we show that the problem is related to a problem in combinatorial set theory, and use this connection to give bounds on the maximum number of APs that a perfectly failure-resistant design with a given number of servelets can support. Our results provide the first rigorous theoretical foundation for practical secure overlay network design.

/pdf/secure-overlay-network-design-bc6zch5f2m.pdf

Secure overlay network design

This paper considers a traditional problem of resource allocation, scheduling jobs on machines. One such recent application is cloud computing, where jobs arrive in an online fashion with capacity requirements and need to be immediately scheduled on physical machines in data centers. It is often observed that the requested capacities are not fully utilized, hence offering an opportunity to employ an overcommitment policy, i.e., selling resources beyond capacity. Setting the right overcommitment level can induce a significant cost reduction for the cloud provider, while only inducing a very low risk of violating capacity constraints. We introduce and study a model that quantifies the value of overcommitment by modeling the problem as a bin packing with chance constraints. We then propose an alternative formulation that transforms each chance constraint into a submodular function. We show that our model captures the risk pooling effect and can guide scheduling and overcommitment decisions. We also develop a family of online algorithms that are intuitive, easy to implement and provide a constant factor guarantee from optimal. Finally, we calibrate our model using realistic workload data, and test our approach in a practical setting. Our analysis and experiments illustrate the benefit of overcommitment in cloud services, and suggest a cost reduction of 1.5% to 17% depending on the provider's risk tolerance.

https://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID3140118_code2182143.pdf?abstractid=2822188&mirid=1&type=2

Vahab Mirrokni

Papers

Parallel and Streaming Algorithms for K-Core Decomposition

Stars: Tera-Scale Graph Building for Clustering and Learning

Pricing against a Budget and ROI Constrained Buyer

Secure overlay network design

Overcommitment in Cloud Services - Bin Packing with Chance Constraints