It is important to detect and handle anomalous inputs when deploying machine learning systems. The use of larger and more complex inputs in deep learning magnifies the difficulty of distinguishing between anomalous and in-distribution examples. At...

We propose DeepV2D, an end-to-end differentiable system which applies both monocular and geometric cues to predict 3D structure from a video sequence. DeepV2D is built from a set of geometric modules based on classical SfM operations and applies s...

In this paper, we have present LDAM, an SGLD-based monte-carlo sampling algorithm that can generate images activating selected neurons in a deep network. In addition, we have discussed some of the pitfalls of current AM methods, which are tending ...

In this paper, we formally defined dynamical walking on slippery surfaces from a hybrid system perspective. This definition made it possible to formally decomposition this problem into a traditional trajectory optimization problem and, for the fir...

In this work, we present a simple method for attacking both image and video classifiers. The proposed attack is universal (i.e. the same adversarial framing can be applied in different images or videos), efficient and effective. Moreover, our meth...

Since the very act of deploying a system can result in shifts that bias the system in practice (e.g., Lum and Isaac (2016)), machine learning practitioners need to become increasingly aware of how deployment and training environments can differ. T...

In this paper, we demonstrated that weighted reference point MMD can closely match the full MMD with significantly less computation and storage. We bound the error in terms of heat diffusion from the reference points, and the bound suggests an opt...

Powerful acquisition function strategies are often computationally intensive and algorithmically complicated [29]. Yet when cast in the space of probabilistic programs, advanced Bayesian optimization techniques can be rather trivial. We demonstrat...

In our plots we show a comparison to BatchNorm and GroupNorm. We omit a comparison to Layer/Instance Normalization as our initial experiments were consistant with findings in the literature that show that they are inferior to GroupNorm and BatchNo...

The negative results presented here are disheartening, however we don’t believe that they rule out the use of stochastic variance reduction on deep learning problems. Rather, they suggest avenues for further research. For instance, SVR can be appl...

This study seeks to use a DNN-based choice model to examine the classical travel mode choice decisions with an emphasis on interpretability. Built upon Wang and Zhao (2018) Wang and Zhao (2018), this paper numerically computes the key economic inf...

We gratefully acknowledge the support of the NIH via grant 1R01HL118181-01A1.

Overall, we conclude that the structure imposed on argumentation by Socratrees accomplishes our goals to make arguments transparent, inspire critical thinking, help find common ground, and be conducive to discussion—at the cost of making it harder...

We tested the evolutionary framework against two problems with known solutions: finding a decomposition of a unitary discrete Fourier transform, which is at the core of Shor’s quantum factoring algorithm shor1999polynomial, and finding Grover’s al...

In this paper, we propose a novel framework which jointly learns a generation network to achieve Chinese glyph synthesis and two encoding networks of the style and content related features using an adversarial process. By the pairwise substitution...

In this study, we apply machine learning algorithms on SMILES, which is a convenient linear graph notation containing information of a molecular structure. It is well known that there are many valid SMILES for a given molecule. The model we propos...

In this paper, we have investigated the challenging problem of deception detection from real-life videos. For joint deep feature learning from facial expressions and body motions, we have proposed a novel face-focused cross-stream network (FFCSN)....

In this paper, we build an inference engine for learning smooth probability functions from a set of Bernoulli experiments, which may be influenced by contextual features. We design an efficient and scalable algorithm for computing a posterior conv...

In this paper, we have investigated the challenging problem of ZSL with less human annotation. For the first time, we have defined a new ZSL setting where only few annotated seen class images are given for training. To overcome the annotation spar...

We would like to express our thanks to Andrea Turkalo, who took the images of the elephants used in this work, and the Elephant Listening Project at the Cornell Lab of Ornithology for preparation of the material and metadata. We would also like to...