This is the final installment of a three-part series exploring the impact of artificial intelligence (AI) on investment management. I want to thank the speakers at the AI and the Future of Financial Services Forum, hosted by CFA Institute and CFA Society Beijing, for inspiring this series. The initial articles offered a primer on the AI technologies that are relevant to investment professionals and explored the potential threat AI posed to human portfolio managers. Not all is lost, investment professionals. Despite artificial intelligence (AI)'s significant and rapidly increasing "brain" power, the investment management business is not going away tomorrow.

"Artificial Intelligence has to be harnessed to find digital innovations for better customer interface and better service delivery. Artificial Intelligence is about creating trains with brains," said Goyal. "Artificial Intelligence can transform Indian Railways in terms of safety, passenger amenities, better revenues, growth and efficiency," he added. Speakers at the conference, organised by Centre for Railway Information System (CRIS), highlighted how the AI can help in enhancing customer experience as well as improve revenues. It was noted that use of AI can help in route optimisation, real-time train movement monitoring, crew fostering, improving the price advantage of rail freight, bettering logistics chain integration, among numerous other applications.

Authorities arrested 26 suspects who used drones to fly two 200-meter (660-feet) cables between Hong Kong and the mainland to transport refurbished iPhones with a total value of 500 million yuan, the paper said in a report on the crackdown by Shenzhen and Hong Kong customs.

Paro the furry seal cries softly while an elderly woman pets it. The upright Tree guides a disabled man taking shaky steps, saying in a gentle feminine voice, "right, left, well done!" Robots have the run of Tokyo's Shintomi nursing home, which uses 20 different models to care for its residents. Allowing them to help care for the elderly -- a job typically seen as requiring a human touch -- may be a jarring idea in the West. But many Japanese see robots positively, largely because they are depicted in popular media as friendly and helpful.

Mizuho and Mitsubishi are speeding up the adoption of artificial intelligence. The two companies have agreed to mutually independent deals with tech firms that will help them go beyond routine banking tasks. Mizuho Financial Group joined IBM to develop a forecasting engine based on artificial intelligence. The engine is imitating the human brain's learning structure to help forecast moves of the financial market, the bank said in a statement. Technically speaking the tool uses artificial neural network technology.

China now has 93 pilot projects involving self-driving cars and shared-driving services nationwide, according to KPMG LLP. Some auto-industry leaders warn that developers may be moving too quickly toward testing on public roads. "It has to be based on solid research--you can't just put some sensors on top of a car," Li Shufu, the chairman of auto maker Zhejiang Geely Holding Group Co., said at an autonomous-driving conference in Beijing on Wednesday. The government should intervene, Mr. Li said, and keep unqualified aspirants off the road. In January, it published a draft blueprint for the country's strategic development of autonomous driving as part of a broader plan to dominate the artificial-intelligence business.

Traffic police in China are to begin using facial-recognition technology to identify jaywalkers and automatically issue them fines by text. Authorities in Shenzhen already publicly name and shame people who flout the southern city's strict road rules, using CCTV cameras equipped with artificial intelligence (AI) that can recognise offenders. Their faces are then displayed on large screens at crossings and on a government website. Now, the company which provides the technology is in talks with mobile phone carriers and social media firms about developing a system that notifies jaywalkers through instant messages when they are caught by the cameras, crossing the road outside of a marked pedestrian crosswalk at an intersection. "Jaywalking has always been an issue in China and can hardly be resolved just by imposing fines or taking photos of the offenders," Wang Jun, director of marketing solutions at Shenzhen-based AI firm Intellifusion, told the South China Morning Post.

The subgraph isomorphism problem involves deciding whether a copy of a pattern graph occurs inside a larger target graph. The non-induced version allows extra edges in the target, whilst the induced version does not. Although both variants are NP-complete, algorithms inspired by constraint programming can operate comfortably on many real-world problem instances with thousands of vertices. However, they cannot handle arbitrary instances of this size. We show how to generate "really hard" random instances for subgraph isomorphism problems, which are computationally challenging with a couple of hundred vertices in the target, and only twenty pattern vertices. For the non-induced version of the problem, these instances lie on a satisfiable / unsatisfiable phase transition, whose location we can predict; for the induced variant, much richer behaviour is observed, and constrainedness gives a better measure of difficulty than does proximity to a phase transition. These results have practical consequences: we explain why the widely researched "filter / verify" indexing technique used in graph databases is founded upon a misunderstanding of the empirical hardness of NP-complete problems, and cannot be beneficial when paired with any reasonable subgraph isomorphism algorithm.

Despite their great success in practical applications, there is still a lack of theoretical and systematic methods to analyze deep neural networks. In this paper, we illustrate an advanced information theoretic methodology to understand the dynamics of learning and the design of autoencoders, a special type of deep learning architectures that resembles a communication channel. By generalizing the information plane to any cost function, and inspecting the roles and dynamics of different layers using layer-wise information quantities, we emphasize the role that mutual information plays in quantifying learning from data. We further propose and also experimentally validate, for mean square error training, two hypotheses regarding the layer-wise flow of information and intrinsic dimensionality of the bottleneck layer, using respectively the data processing inequality and the identification of a bifurcation point in the information plane that is controlled by the given data. Our observations have direct impact on the optimal design of autoencoders, the design of alternative feedforward training methods, and even in the problem of generalization.

Modes and ridges of the probability density function behind observed data are useful geometric features. Mode-seeking clustering assigns cluster labels by associating data samples with the nearest modes, and estimation of density ridges enables us to find lower-dimensional structures hidden in data. A key technical challenge both in mode-seeking clustering and density ridge estimation is accurate estimation of the ratios of the first- and second-order density derivatives to the density. A naive approach takes a three-step approach of first estimating the data density, then computing its derivatives, and finally taking their ratios. However, this three-step approach can be unreliable because a good density estimator does not necessarily mean a good density derivative estimator, and division by the estimated density could significantly magnify the estimation error. To cope with these problems, we propose a novel estimator for the \emph{density-derivative-ratios}. The proposed estimator does not involve density estimation, but rather \emph{directly} approximates the ratios of density derivatives of any order. Moreover, we establish a convergence rate of the proposed estimator. Based on the proposed estimator, novel methods both for mode-seeking clustering and density ridge estimation are developed, and the respective convergence rates to the mode and ridge of the underlying density are also established. Finally, we experimentally demonstrate that the developed methods significantly outperform existing methods, particularly for relatively high-dimensional data.