Abstract-Deep neural networks have achieved near-human accuracy levels in various types of classification and prediction tasks including images, text, speech, and video data. However, the networks continue to be treated mostly as black-box function approximators, mapping a given input to a classification output. The next step in this human-machine evolutionary processincorporating these networks into mission critical processes such as medical diagnosis, planning and control -requires a level of trust association with the machine output.Typically, statistical metrics are used to quantify the uncertainty of an output. However, the notion of trust also depends on the visibility that a human has into the working of the machine. In other words, the neural network should provide humanunderstandable justifications for its output leading to insights about the inner workings. We call such models as interpretable deep networks.Interpretability is not a monolithic notion. In fact, the subjectivity of an interpretation, due to different levels of human understanding, implies that there must be a multitude of dimensions that together constitute interpretability. In addition, the interpretation itself can be provided either in terms of the lowlevel network parameters, or in terms of input features used by the model. In this paper, we outline some of the dimensions that are useful for model interpretability, and categorize prior work along those dimensions. In the process, we perform a gap analysis of what needs to be done to improve model interpretability.

No abstract

The preparation is described of the trifluoromethoxides of the heavier alkali metals, i.e. MOCF, where M = K, Rb, or Cs. These are stable, crystalline, ionic solids made by the reversible reaction of carbonyl fluoride with the appropriate metal fluoride. No trifluoromethoxides could be prepared from lithium, sodium, barium, or thallium(1) fluorides. Factors affecting the relative stability of such compounds are discussed.Trifluoromethanol, CFBOH, is not known, and is generally assumed to be incapable of existence because of the ease with which compounds having a fluorine atom in a position a to a hydroxyl group would eliminate hydrogen fluoride. Because of this unavailability of the parent alcohol, no metallic trifluoromethoxides have hitherto been reported. We now report the preparation of stable, crystalline trifluoromethoxides of the heavier alkali metals, i.e. MOCF3, where RII = K, Rb, or Cs.A consideration of the possible formation of metallic trifluoromethoxides shows a t once that they are likely to be more ionic in nature than the corresponding methoxides. The completely ionic formulation M+[OCFJd would contain the trifluoromethoxide ion, isoelectronic with the fluoroborate ion, BF4-. The charge in the trifluoromethoxide ion rests formally on the oxygen atom, but it would be shared to some extent by the three fluorine atoms because of their greater electronegativity. The distribution of charge over the whole ion achieved in this way should therefore stabilize it, in a manner not possible in the ionic non-fluorinated methoxides of the alkali metals. The stability of a solid ionic trifluoromethoxide would clearly be influenced by the nature of the cation present, since smaller cations, or those with a high ratio of charge to size, would cause greater polarization of the anion and a correspondingly greater tendency towards decomposition to give a fluoride ion and carbonyl fluoride. In other words, the equilibrium:would be displaced to the left by the presence of large cations. An immediate corollary of this suggestion is that by approaching the equilibrium from the direction of formation rather than decomposition one would have available a possible synthetic route to trifluoromethoxides.

No abstract

No abstract