Saddle Points and Accelerated Perceptron Algorithms

Authors: Adams Wei Yu, Fatma Kilinc-Karzan, Jaime Carbonell

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We support our theoretical findings with an empirical study on synthetic and real data, highlighting the efficiency and numerical stability of our algorithm, especially on large-scale instances.
Researcher Affiliation Academia Adams Wei Yu WEIYU@CS.CMU.EDU Fatma Kılınc -Karzan FKILINC@ANDREW.CMU.EDU Jaime G. Carbonell JGC@CS.CMU.EDU School of Computer Science, Carnegie Mellon University, Pittsburgh, PA, USA Tepper School of Business, Carnegie Mellon University, Pittsburgh, PA, USA
Pseudocode Yes Algorithm 1 MPFP 1: Input: ω-center zω, step size {γt} and ϵ. 2: Output: zt(= [xt; yt]). 3: t = 1; v1 = zω; 4: while φ(yt) 0 and ϵsad(zt) > ϵ do 5: wt = Proxvt(γt F(vt)); 6: vt+1 = Proxvt(γt F(wt)); 7: zt = h Pt s=1 γs i 1 Pt s=1 γsws; 8: t = t + 1; 9: end while
Open Source Code No The paper states that 'All the codes are written in Matlab 2011b', but it does not provide an explicit statement about the public availability of the source code or a link to a repository.
Open Datasets Yes Our testbed is the CIFAR-10 (Krizhevsky, 2009) dataset. [...] We evenly pick around 1,000 images in each class to construct a training set of n = 10, 000 in total, and repeat this process to form a testing set of the same size.
Dataset Splits No The paper mentions a training set and a testing set, but it does not specify a separate validation set or describe a cross-validation strategy.
Hardware Specification Yes All the codes are written in Matlab 2011b, and are ran in a single-threaded fashion on a Linux server with 6 dual core 2.8GHz CPU and 64 GB memory.
Software Dependencies Yes All the codes are written in Matlab 2011b
Experiment Setup Yes In the implementation of MPLFP or MPKFP, we select αx, αy as described in the appendix, for a normalized matrix A, i.e., Aj 2 = 1 for all j. [...] The iteration limit is 106 and runtime limit is 5 104s. [...] We choose Radial Basis Function (RBF) kernel given by KΦ(Ai, Aj) = exp{ 5.5( Ai Aj 2)}.