The ICML 2013 Workshop on Challenges in Representation Learning(1) focused on three challenges: the black box learning challenge, the facial expression recognition challenge, and the multimodal learning challenge. We describe the datasets created for these challenges and summarize the results of the competitions. We provide suggestions for organizers of future challenges and some comments on what kind of knowledge can be gained from machine learning competitions.

Abstract. The ICML 2013 Workshop on Challenges in Representation Learning3 focused on three challenges: the black box learning challenge, the facial expression recognition challenge, and the multimodal learning challenge. We describe the datasets created for these challenges and summarize the results of the competitions. We provide suggestions for organizers of future challenges and some comments on what kind of knowledge can be gained from machine learning competitions.

With a recently developed unique deep ultraviolet picoseconds time-resolved photoluminescence (PL) spectroscopy system and improved growth technique, we are able to determine the detailed band structure near the Γ point of wurtzite (WZ) AlN with a direct band gap of 6.12 eV. Combined with first-principles band structure calculations we show that the fundamental optical properties of AlN differ drastically from that of GaN and other WZ semiconductors. The discrepancy in energy band gap values of AlN obtained previously by different methods is explained in terms of the optical selection rules in AlN and is confirmed by measurement of the polarization dependence of the excitonic PL spectra.

Using a band-structure method that includes the correction to the band-gap error in the local-density approximation (LDA), we find that the band gap for InN is 0.8±0.1eV, in good agreement with recent experimental data, but is much smaller than previous experimental value of ∼1.9eV. The unusually small band gap for InN is explained in terms of the high electronegativity of nitrogen and, consequently, the small band-gap deformation potential of InN. The possible origin of the measured large band gaps is discussed in terms of the nonparabolicity of the bands, the Moss–Burstein shift, and the effect of oxygen. Based on the error analysis of our LDA-corrected calculations we have compiled the band-structure parameters for wurtzite AlN, GaN, and InN.

We present a general scheme for treating the integrable singular terms within exact exchange (EXX) Kohn-Sham or Hartree-Fock (HF) methods for periodic solids. We show that the singularity corrections for treating these divergencies depend only on the total number and the positions of k points and on the lattice vectors, in particular the unit cell volume, but not on the particular positions of atoms within the unit cell. The method proposed here to treat the singularities constitutes a stable, simple to implement, and general scheme that can be applied to systems with arbitrary lattice parameters within either the EXX Kohn-Sham or the HF formalism. We apply the singularity correction to a typical symmetric structure, diamond, and to a more general structure, trans-polyacetylene. We consider the effect of the singularity corrections on volume optimisations and k point convergence. While the singularity corrections clearly depends on the total number of k points, it exhibits a remarkably small dependence upon the choice of the specific arrangement of the k points.