Compressing to VC Dimension Many Points
نویسنده
چکیده
Note that the VC dimension of orthogonal rectangles is four and this is exactly the number of examples needed to represent the consistent orthogonal rectangle. A compression scheme of size k for a concept class C picks from any set of examples consistent with some concept in C a subset of up to k examples and this subset represents (via a mapping that that is specific to the class C) a hypothesis consistent with the whole original set of examples. Conjecture: Any concept class of VC dimension d has a compression scheme of size d. What evidence do we have that this conjecture might be true? Call a concept class of VC dimension d maximum if for every subset of m instances, the concept class induces exactly ∑ d
منابع مشابه
VC-Dimension of Visibility on Terrains
A guarding problem can naturally be modeled as a set system (U ,S) in which the universe U of elements is the set of points we need to guard and our collection S of sets contains, for each potential guard g, the set of points from U seen by g. We prove bounds on the maximum VC-dimension of set systems associated with guarding both 1.5D terrains (monotone chains) and 2.5D terrains (polygonal ter...
متن کاملExact VC-dimension for $L_1$-visibility of points in simple polygons
The VC-dimension plays an important role for the algorithmic problem of guarding art galleries efficiently. We prove that inside a simple polygon at most 5 points can be shattered by L1-visibility polygons and give an example where 5 points are shattered. The VCdimension is exactly 5. The proof idea for the upper bound is different from previous approaches.
متن کامل0.1 Outline
A lower bound on sample complexity through VC-dimension. Cover's coin problem. 10.2 A lower bound on the sample complexity through VC-dimension In previous lectures it has been shown that the VC-dimension of a concept class gives an upper bound on the number of samples needed to PAC-learn concepts from the class. In particular it has been shown that for VC-dimension d it is possible to PAC-lear...
متن کاملTeaching and compressing for low VC-dimension
In this work we study the quantitative relation between VC-dimension and two other basic parameters related to learning and teaching. We present relatively efficient constructions of sample compression schemes and teaching sets for classes of low VC-dimension. Let C be a finite boolean concept class of VC-dimension d. Set k = O(d2d log log |C|). We construct sample compression schemes of size k...
متن کاملOn Interference Among Moving Sensors and Related Problems
We study geometric hypergrahs in a kinetic setting. That is, the set of vertices of the hypergraph is a set of moving points in R with coordinates that are polynomials in time. The hyperedges are all subsets that can be realized by intersecting the set of points at some fixed time with some “simple” geometric shape, such as, say, a halfspace. We show that for many of the static cases where the ...
متن کامل