The historical improvements in the performance of general-purpose processors have long provided opportunities for application innovation. Word processing, spreadsheets, desktop publishing, networking and various game genres are just some of the many applications that have arisen because of the increasing capabilities and the versatility of general-purpose processors. Key to these innovations is the fact that general-purpose processors do not predefine the applications that they are going to run.

Currently, the capabilities of individual general-purpose processors are encountering challenges, such as diminishing returns in exploiting instruction-level parallelism and power limits. As a consequence, a variety of approaches are being employed to address this situation, including adding myriad dedicated accelerators. Unfortunately, while this improves performance it sacrifices generality. More specifically, the time, difficulty and cost of special purpose design preclude dedicated logic from serving as a viable avenue for application innovation.

There recently has been progress in addressing this dilemma between providing programmability and higher performance via an interesting middle ground between fully general-purpose computing and dedicated logic. In specific, spatial computing, where the computation is mapped spatially onto an array of small programmable processing elements addresses many of the cost-related liabilities of dedicated logic and is increasingly being applied to general computation problems. While field prgrammable gate arrays (FPGAs) are the best know spatial computing platform there are also a number of coarse grained variants.

In this talk, we will examine the range of spatial computing alternatives and explore in more depth the concept of triggered instructions, a novel control paradigm for arrays of processing elements (PEs) aimed at exploiting spatial parallelism. Triggered instructions completely eliminate the program counter and allow programs to transition concisely between states without explicit branch instructions. They also allow efficient reactivity to inter-PE communication traffic. The approach provides a unified mechanism to avoid over-serialized execution, essentially achieving the effect of techniques such as dynamic instruction reordering and multithreading, which each require distinct hardware mechanisms in a traditional sequential architecture.

The dream of machines that prove mathematical theorems automatically goes back to the beginning computer science. Meanwhile it has become a reality, although we had to replace "automatic" with "interactive" to make it work. Modern Interactive Theorem Provers (or Proof Assistants) are like programming environments, except that they help us write proofs instead of programs. Crucially, they check the proofs for logical correctness and (try to) bridge gaps in the argument by constructing missing subproofs automatically.
This talk will present an overview of the field from the perspective of the Isabelle proof assistant. I will talk about a range of applications from the proof of the Kepler Conjecture via the analysis of programming languages and operating systems to the role of proof assistants in teaching. A demo of the Isabelle systems will help the audience to get a feeling for what is involved when trying to convince a machine that some theorem is true.

How can people and machines cooperate to gain insight and discover useful information from complex data sets?  A central challenge lies in optimizing the interaction, which leads to difficult sequential decision problems under uncertainty.  In this talk, I will introduce the new concept of adaptive submodularity, generalizing the classical notion of submodular set functions to adaptive policies. We prove that if a problem satisfies this property, a simple adaptive greedy algorithm is guaranteed to be competitive with the optimal policy.
The concept allows to recover, generalize and extend existing results in diverse domains including active learning, resource allocation and social network analysis. I will show results on several real-world applications, ranging from interactive content search over image categorization in citizen science to biodiversity monitoring via conservation drones.

People have always been influenced by the opinions of their acquaintances.  Increasingly, through recommendations and ratings provided on all sorts of goods and services, people are also influenced by the opinions of people that are not even acquaintances.  This ubiquity of the sharing of opinions has intensified the interest is the concept of herding (or informational cascades) introduced in 1992.  While agents in most previous works have only individualistic goals, this talk focuses on social influence among agents in two collaborative settings.

We consider agents that perform Bayesian binary hypothesis testing and, in addition to their private signals, observe the decisions of earlier-acting agents.  In the first setting, each decision has its own corresponding Bayes risk.  Each agent affects the minimum possible Bayes risk for subsequent agents, so an agent may have a mixed objective including her own Bayes risk and the Bayes risks of subsequent agents; we demonstrate her tension between being informative to other agents and being right in her own decisions, and we show that she is more informative to others when she is open minded.  In the second setting, opinions are aggregated by voting, and all agents aim to minimize the Bayes risk of the team's decision.  We show that social learning is futile when the agents observe conditionally independent and identically distributed private signals (but not merely conditionally independent private signals) or when the agents require unanimity to make a decision.  Our experiments with human subjects suggest that when opinions of people with equal qualities of information are aggregated by voting, the ballots should be secret.  They have also raised questions about rationality and trust.

BitCoin is a digital currency system introduced in 2008 by an anonymous developer using a pseudonym "Satoshi Nakamoto".  Despite of its mysterious origins, Bitcoin became the first cryptographic currency that got widely adopted --- as of May 2014 the Bitcoin capitalization is over 5 bln euro.  Bitcoin owes its popularity mostly to the fact that it has no central authority, the transaction fees are very low, and the amount of coins in the circulation is restricted, which in particular means that nobody can "print" money to generate inflation. The financial transactions between the participants are published on a public ledger maintained jointly by the users of the system. 

One of the very interesting, but slightly less known, features of the Bitcoin is the fact that it allows for more complicated "transactions" than the simple money transfers between the participants: very informally, in Bitcoin it is possible to "deposit" some amount of money in such a way that it can be claimed only under certain conditions.  These conditions  are written in the form of the "Bitcoin scripts" and in particular may involve some timing constrains.  This property allows to create the so-called "contracts", where a number of mutually-distrusting parties engage in a Bitcoin-based protocol to jointly perform some task. The  security of the protocol is guaranteed purely by the properties of the Bitcoin, and no additional trust assumptions are needed.  This Bitcoin feature can have several applications in the digital economy, like creating the assurance contracts, the escrow and dispute mediation, the rapid micropayments, the multiparty lotteries.

In this talk I will give a short introduction to this area, present some recent results, and highlight the future research directions.

The end of Dennard scaling is expected to shrink the range of DVFS in
future nodes, limiting the energy savings of this technique. This work
evaluates how much we can increase the effectiveness of DVFS by using a
software decoupled access-execute approach. Decoupling the data access
from execution allows us to apply optimal voltage-frequency selection
for each phase and therefore improve energy efficiency over standard
coupled execution.
The underlying insight of our work is that by decoupling access and
execute we can take advantage of the memory-bound nature of the access
phase and the compute-bound nature of the execute phase to optimize
power efficiency, while maintaining good performance. To demonstrate
this we built a task based parallel execution infrastructure consisting
of a runtime system to orchestrate the execution, compiler
infrastructure to automatically concert programs to
Decoupled-Access-Execute, and a modeling infrastructure based on
hardware measurements to simulate zero-latency, per-core DVFS.
Based on real hardware measurements we project that the combination of
decoupled access-execute compiled programs and DVFS has the potential to
improve EDP by 25% without hurting performance. On memory-bound
applications we significantly improve performance due to increased MLP
in the access phase and ILP in the execute phase. Furthermore we
demonstrate that our method can achieve high performance both in
presence or absence of a hardware prefetcher. We are now expanding our
work to target serial programs using a combination of compiler
techniques and hardware features.

The volume and surface scattering of a material is usually defined at a macroscopic scale by its bidirectional reflectance distribution function (BRDF) that describes the chance of reflection for different pairs of incoming and outgoing light directions. It is widely used in remote sensing, science material, lighting simulation or computer graphics.
Microfacet BRDF models are based on a geometrical description of the material surface as a collection of microfacets whose dimensions are much greater than the wavelength. The resulting BRDF dépends on the microfacet slope distribution, the shadowing-masking function and the BRDF of the individual facets. In the original Cook-Torrance model, and in most of cases even today, each facet is assumed to reflect light only in the specular direction. The third factor of the model is then given by the Fresnel relation depending on the material refractive index.
We will present the generalization of microfacet models for any radiometric response of the individual facets. Some crucial points (radiometric proof of the model, choice of the distribution and of the shadowing-masking function, numerical calculations…) will be discussed. And we propose to consider facets consisting of a flat interface on a Lambertian background. The Oren-Nayar model distribution of strictly Lambertian facets and the Cook-Torrance model distribution of strictly specular facets appear as special cases.