Prototype-based classification learning methods are known to be inherently interpretable. However, this paradigm suffers from major limitations compared to deep models, such as lower performance. This led to the development of the so-called deep Prototype-Based Networks (PBNs), also known as prototypical parts models. In this work, we analyze these models with respect to different properties, including interpretability. In particular, we focus on the Classification-by-Components (CBC) approach, which uses a probabilistic model to ensure interpretability and can be used as a shallow or deep architecture. We show that this model has several shortcomings, like creating contradicting explanations. Based on these findings, we propose an extension of CBC that solves these issues. Moreover, we prove that this extension has robustness guarantees and derive a loss that optimizes robustness. Additionally, our analysis shows that most (deep) PBNs are related to (deep) RBF classifiers, which implies that our robustness guarantees generalize to shallow RBF classifiers. The empirical evaluation demonstrates that our deep PBN yields state-of-the-art classification accuracy on different benchmarks while resolving the interpretability shortcomings of other approaches. Further, our shallow PBN variant outperforms other shallow PBNs while being inherently interpretable and exhibiting provable robustness guarantees.

Neural networks are not great generalizers outside their training range i.e. they are good at capturing bias but might miss the overall concept. Important issues with neural networks is that when testing data goes outside training range they fail to predict accurate results. Hence, they loose the ability to generalize a concept. For systematic numeric exploration neural accumulators (NAC) and neural arithmetic logic unit(NALU) are proposed which performs excellent for simple arithmetic operations. But, major limitation with these units is that they can't handle complex mathematical operations & equations. For example, NALU can predict accurate results for multiplication operation but not for factorial function which is essentially composition of multiplication operations only. It is unable to comprehend pattern behind an expression when composition of operations are involved. Hence, we propose a new neural network structure effectively which takes in complex compositional mathematical operations and produces best possible results with small NALU based neural networks as its pluggable modules which evaluates these expression at unitary level in a bottom-up manner. We call this effective neural network as CalcNet, as it helps in predicting accurate calculations for complex numerical expressions even for values that are out of training range. As part of our study we applied this network on numerically approximating complex equations, evaluating biquadratic equations and tested reusability of these modules. We arrived at far better generalizations for complex arithmetic extrapolation tasks as compare to both only NALU layer based neural networks and simple feed forward neural networks. Also, we achieved even better results for our golden ratio based modified NAC and NALU structures for both interpolating and extrapolating tasks in all evaluation experiments.Finally, from reusability standpoint this model demonstrate strong invariance for making predictions on different tasks.