We also propose a new algorithm that learns to estimate metric depth using annotations of relative depth. Compared to the state of the art, our algorithm is simpler and performs better.

This paper studies single-image depth perception in the wild, i.e., recovering depth from a single image taken in unconstrained settings. We introduce a new dataset "Depth in the Wild" consisting of images in the wild annotated with relative depth between pairs of random points. We also propose a new algorithm that learns to estimate metric depth using annotations of relative depth. Compared to the state of the art, our algorithm is simpler and performs better. Experiments show that our algorithm, combined with existing RGB-D data and our new relative depth annotations, significantly improves single-image depth perception in the wild.

The behaviors of deep neural networks (DNNs) are notoriously resistant to human interpretations. In this paper, we propose Hypergradient Data Relevance Analysis, or HYDRA, which interprets the predictions made by DNNs as effects of their training data. Existing approaches generally estimate data contributions around the final model parameters and ignore how the training data shape the optimization trajectory. By unrolling the hypergradient of test loss w.r.t. the weights of training data, HYDRA assesses the contribution of training data toward test data points throughout the training trajectory. In order to accelerate computation, we remove the Hessian from the calculation and prove that, under moderate conditions, the approximation error is bounded. Corroborating this theoretical claim, empirical results indicate the error is indeed small. In addition, we quantitatively demonstrate that HYDRA outperforms influence functions in accurately estimating data contribution and detecting noisy data labels. The source code is available at https://github.com/cyyever/aaai_hydra_8686.

We present SymForce, a library for fast symbolic computation, code generation, and nonlinear optimization for robotics applications like computer vision, motion planning, and controls. SymForce combines the development speed and flexibility of symbolic math with the performance of autogenerated, highly optimized code in C++ or any target runtime language. SymForce provides geometry and camera types, Lie group operations, and branchless singularity handling for creating and analyzing complex symbolic expressions in Python, built on top of SymPy. Generated functions can be integrated as factors into our tangent-space nonlinear optimizer, which is highly optimized for real-time production use. We introduce novel methods to automatically compute tangent-space Jacobians, eliminating the need for bug-prone handwritten derivatives. This workflow enables faster runtime code, faster development time, and fewer lines of handwritten code versus the state-of-the-art. Our experiments demonstrate that our approach can yield order of magnitude speedups on computational tasks core to robotics. Code is available at https://github.com/symforce-org/symforce.

Why do we sometimes perceive static images as if they were moving? Visual motion illusions enjoy a sustained popularity, yet there is no definitive answer to the question of why they work. We present a generative model, the Evolutionary Illusion GENerator (EIGen), that creates new visual motion illusions. The structure of EIGen supports the hypothesis that illusory motion might be the result of perceiving the brain's own predictions rather than perceiving raw visual input from the eyes. The scientific motivation of this paper is to demonstrate that the perception of illusory motion could be a side effect of the predictive abilities of the brain. The philosophical motivation of this paper is to call attention to the untapped potential of "motivated failures", ways for artificial systems to fail as biological systems fail, as a worthy outlet for Artificial Intelligence and Artificial Life research.

Natural Language Processing (NLP), the ability of a software program to understand human language as it is spoken, has seen major breakthroughs, thanks to Artificial Intelligence (AI) and improved access to fast processors and cloud computing. With the introduction of more personal assistants, better smartphone functionality, and the evolution of Big Data to automate even more regular human jobs, NLP adoption is projected to gain up steam in the future years. SoundHound creates AI and conversational intelligence systems that are voice-enabled. It offers a Speech-to-Meaning engine as well as Deep Meaning Understanding technology, which may be integrated into other services and devices. It also creates music recognition apps and voice search assistants.

A perfect storm of pressures which has been building on the financial services industry for years has come to a head. Covid-19 has been a catalyst for change for institutions who have been facing a build-up of competitive and regulatory pressures and now have to face the urgency of making their business models fit for purpose. AI, for many years a "nice to have", has now become integral to running a financial services business efficiently and profitably. US banks are ahead of the curve here: many have already gone through their AI transformation and as a result are in better shape for the next decade. Many European institutions need their own AI revolution in order not to become obsolete.

One of the bigger trends in enterprise software has been the emergence of startups building tools to make the benefits of artificial intelligence technology more accessible to non-tech companies. Today, one that has built a platform to apply power of machine learning and natural language processing to massive documents of unstructured data has closed a round of funding as it finds strong demand for its approach. Eigen Technologies, a London-based startup whose machine learning engine helps banks and other businesses that need to extract information and insights from large and complex documents like contracts, is today announcing that it has raised $37 million in funding, a Series B that values the company at around $150 million – $180 million. The round was led by Lakestar and Dawn Capital, with Temasek and Goldman Sachs Growth Equity (which co-led its Series A) also participating. Eigen has now raised $55 million in total.

Hiscox has teamed up with tech startup Eigen Technologies to speed up how internal data is processed and reduce costs. Two pilots are underway to test the suitability of Eigen's NLP (natural language processing) platform at addressing the challenges Hiscox faces with unstructured qualitative data. It will explore machine learning (ML) and extraction technology. Currently the global specialist insurer and underwriter for Lloyd's London relies heavily on manual data entry. It hopes to transform how data is managed.

Readers will probably have heard of Eigen and the work it is doing in the legal AI doc review space, but perhaps have not had a chance to hear its inspiring founder speak. Here is a short video presentation (13.27 mins) recorded at a recent Codex.com'So