Miklos Ajtai
The role of lattices in complexity theory and cryptography:
There are many ways of generating lattices at random. For some of
these randomizations
there are equivalences between average-case and worst-case
algorithmic problems. Such
an equivalence can be used for cryptographic purposes. The reason why
such an equivalence
exists for lattices is the symmetricity (that is, invariance under
linear transformations) inherent in
the set of all lattices and in the distribution that we use for
picking a random lattice. This distribution
sometimes only approximates a symmetric one so it may not be
apparent. There is another symmetry
in the possible ways of the representations of a lattice by different
bases. These symmetries imply
the nice properties of lattices which make average-case worst-case
equivalences possible.
Shuchi Chawla
Approximation Algorithms for Planning Problems:
Given a weighted undirected graph with "rewards" on nodes, the
Orienteering problem involves finding a path of length at most D in the
graph between two given nodes that covers the maximum possible reward.
Such "path-planning" problems involving timing constraints arise in a
variety of applications, for example, transportation and delivery
problems, supply chain management, and robot navigation. We consider a
variety of path-planning problems with timing constraints, and review some
recent results on approximating these problems. We also consider planning
problems on probabilistic graphs (Markov decision processes), and present
some preliminary results for them.
Michael Luby
Application-layer FEC codes for mobile wireless broadcast and multicast
services:
Abstract TBA.
Constantinos Daskalakis
The Complexity of Nash Equilibria:
In 1951, Nash showed that every game has a mixed Nash equilibrium. His proof
is a reduction to Brouwer's fixpoint theorem and places the problem of
finding equilibria into the realm of 'exponential existence proofs'. In
fact, whether Nash equilibria can be computed in polynomial time has
remained open since that time, and has come under increased scrutiny during
the past two decades. In this talk, I will present some recent results
(joint work with Paul Goldberg and Christos Papadimitriou) that shed light
to this problem, in the negative direction.
Jason Hartline
Mechanism Design via Machine Learning:
We consider the problem of designing nearly optimal mechanisms without prior
knowledge of the distribution from which agent's valuations are drawn. In
this setting, we use techniques from sample-complexity in machine learning
to reduce problems in optimal mechanism design to those in classical
(non-game theoretic) optimization. Our reductions imply that for these
problems, given an optimal algorithm for the classical optimization problem,
we can convert it into a $(1+\epsilon)$-approximation for the
incentive-compatible mechanism design problem, so long as the number of
bidders is sufficiently large as a function of an appropriate measure of
complexity of the comparison class of solutions. We apply these results to
the problem of auctioning a digital good, the attribute auction problem, and
to the problem of item-pricing in unlimited-supply combinatorial auctions.
From a machine learning perspective, these settings present several
challenges: in particular, the loss function is discontinuous and
asymmetric, and the range of bidders' valuations may be large.
This is joint work with Maria-Florina Balcan, Avrim Blum, and Yishay
Mansour.
Vladlen Koltun
The Arrangement Method for Linear Programming:
We present a new combinatorial method for linear programming. Like
the simplex method, it takes a walk on a graph. However, the graph it walks
on has considerably better structure than polytope graphs, including a small
polynomial diameter. Among other things, this implies that the status of the
Hirsch conjecture can no longer be viewed as a fundamental stumbling block
on the road to a strongly polynomial algorithm for linear programming.
Satish Rao
Recent progress on Sparsest Cuts and Metric Embeddings:
We present ideas from recent results on embedding $\ell_2^2$ metrics
into $\ell_2$ metrics. In particular, we will present the proof of
Arora, Lee and Naor that shows that this is possible with distortion
at most $O^*(\sqrt{\log n})$ (ignoring $\log \log n$ factors.) We start
with building blocks from Arora, Rao, and Vazirani, and Chawla,
Raecke, and Gupta. These results have application in approximation
algorithms and have application in functional analysis.