Recently, Temporal Graph Neural Networks (TGNNs) have demonstrated state-of-the-art performance in various high-impact applications, including fraud detection and content recommendation. Despite the success of TGNNs, they are prone to the prevalent noise found in real-world dynamic graphs like time-deprecated links and skewed interaction distribution. The noise causes two critical issues that significantly compromise the accuracy of TGNNs: (1) models are supervised by inferior interactions, and (2) noisy input induces high variance in the aggregated messages. However, current TGNN denoising techniques do not consider the diverse and dynamic noise pattern of each node. In addition, they also suffer from the excessive mini-batch generation overheads caused by traversing more neighbors. We believe the remedy for fast and accurate TGNNs lies in temporal adaptive sampling. In this work, we propose TASER, the first adaptive sampling method for TGNNs optimized for accuracy, efficiency, and scalability. TASER adapts its mini-batch selection based on training dynamics and temporal neighbor selection based on the contextual, structural, and temporal properties of past interactions. To alleviate the bottleneck in mini-batch generation, TASER implements a pure GPU-based temporal neighbor finder and a dedicated GPU feature cache. We evaluate the performance of TASER using two state-of-the-art backbone TGNNs. On five popular datasets, TASER outperforms the corresponding baselines by an average of 2.3% in Mean Reciprocal Rank (MRR) while achieving an average of 5.1x speedup in training time.

Given its effectiveness on knowledge-intensive natural language processing tasks, dense retrieval models have become increasingly popular. Specifically, the de-facto architecture for open-domain question answering uses two isomorphic encoders that are initialized from the same pretrained model but separately parameterized for questions and passages. This bi-encoder architecture is parameter-inefficient in that there is no parameter sharing between encoders. Further, recent studies show that such dense retrievers underperform BM25 in various settings. We thus propose a new architecture, Task-aware Specialization for dense Retrieval (TASER), which enables parameter sharing by interleaving shared and specialized blocks in a single encoder. Our experiments on five question answering datasets show that TASER can achieve superior accuracy, surpassing BM25, while using about 60% of the parameters as bi-encoder dense retrievers. In out-of-domain evaluations, TASER is also empirically more robust than bi-encoder dense retrievers. Our code is available at https://github.com/microsoft/taser.

Axon has paused work on a project to build drones equipped with its Tasers. A majority of its artificial intelligence ethics board quit after the plan was announced last week. Nine of the 12 members said in a resignation letter that, just a few weeks ago, the board voted 8-4 to recommend that Axon shouldn't move forward with a pilot study for a Taser-equipped drone concept. "In that limited conception, the Taser-equipped drone was to be used only in situations in which it might avoid a police officer using a firearm, thereby potentially saving a life," the nine board members wrote. They noted Axon might decline to follow that recommendation and were working on a report regarding measures the company should have in place were it to move forward.

Summary: Deep learning and Big Data are being adopted in law enforcement and criminal justice at an unprecedented rate. Does this scare you or make you feel safe? When you read the title, whether your mind immediately went for the upstairs "H" or the downstairs "H" probably says something about whether the new applications of Big Data in law enforcement let you sleep like a baby or keep you up at night. You might have thought your choice of "H" related to whether you've been on the receiving end of Big Data in law enforcement but the fact is that practically all of us have, and for those who haven't it won't take much longer to reach you. There is an absolute explosion in the use of Big Data and predictive analytics in our legal system today driven by the latest innovations in data science and by some obvious applications.

After announcing that it would shift some of its emphasis away from non-lethal weapons to police body cameras, for a fleeting moment it felt like the company synonymous with sticks that electrocute people was showing an interest in police accountability. Analysis from the Intercept and a 2017 "Law Enforcement Technology Report" by Taser suggest that the reality might be more complicated -- and considerably creepier. The company now known as Axon created its body camera division a few years ago, but ramped up efforts in 2017. After acquiring two AI companies, Dextro and Fossil Group, in February, signs point to the fact that the company wants to aim its new machine learning brainpower at policing. While the company has explicitly denied its interest in building a predictive policing engine, claiming that it "will not make predictions on behalf of our customers," the industry report makes plain reference to its desire to "automate the collection and analysis of virtually all information in public safety while extracting key insights never before possible." In a page on AI and machine learning, the report lauds the superior insight culled from massive data sets that companies in other industries leverage to predict customer behavior.

With the advent and popularity of video content gaining giant strides by each day, the demand for need to make video content search enabled is also increasing. The overall task is simple – creating machine readable semantic metadata of the videos that can be analyzed using text mining techniques. But this task is a very challenging one. Not only does it require processing of video content at scale but also the preferred approach of breaking down videos into still frames, aka images, has its own challenges. The biggest one being processing 30 frames per second is a trash intensive process that certainly demands lookout for better approaches.

When civil liberties advocates discuss the dangers of new policing technologies, they often point to sci-fi films like "RoboCop" and "Minority Report" as cautionary tales. In "RoboCop," a massive corporation purchases Detroit's entire police department. After one of its officers gets fatally shot on duty, the company sees an opportunity to save on labor costs by reanimating the officer's body with sleek weapons, predictive analytics, facial recognition, and the ability to record and transmit live video. Although intended as a grim allegory of the pitfalls of relying on untested, proprietary algorithms to make lethal force decisions, "RoboCop" has long been taken by corporations as a roadmap. And no company has been better poised than Taser International, the world's largest police body camera vendor, to turn the film's ironic vision into an earnest reality.

Non-lethal weapons manufacturer Taser, also one of the leading body camera developers, announced last week that they are forming an artificial intelligence division within the company. Taser acquired two companies focused on artificial intelligence, Dextro and Misfit, hoping to create powerful surveillance tools that will allow police officers to scan people and objects both in real-time and retroactively. The company hopes to make it easier for law enforcement to scan hours of footage using keywords to zoom in on relevant images, such as searching for a gun or a specific car. Taser claims, however, that the software will not have facial-recognition capabilities just yet, but will simply detect faces for the purpose of redacting them from videos. "To clarify, Dextro's system offers computer vision techniques to identify faces and other objects for improving the efficiency of the redaction workflow. AI enables you to become more targeted when needed," Taser vice president of communications Steve Tuttle, told Vocativ.

The police body camera industry is the latest to jump on the artificial intelligence bandwagon, bringing new powers and privacy concerns to a controversial technology bolstered by the need to hold police accountable after numerous high-profile killings of unarmed black citizens. Now, that tech is about to get smarter. Last week, Taser, the stun gun company that has recently become an industry leader in body-mounted cameras, announced the creation of its own in-house artificial intelligence division. The new unit will utilize the company's acquisition of two AI-focused firms: Dextro, a New York-based computer vision startup, and Misfit, another computer vision company previously owned by the watch manufacturer Fossil. Taser says the newly formed division will develop AI-powered tech specifically aimed at law enforcement, using automation and machine learning algorithms to let cops search for people and objects in video footage captured by on-body camera systems.

Law enforcement agencies across the country are adopting body-worn cameras as a means both of increasing their transparency with the public and generating actionable feedback to improve officer performance. Problem is, all these body cams produce terabytes of data daily, far more than many departments can effectively handle. That's why Taser (yes those guys, they make body cameras too) announced on Thursday that it has acquired a pair of companies that specialize in computer- and machine-vision to create the "Axon AI" group. Together, they'll develop a platform that can efficiently parse this flood of data in real time. The Axon AI group will include about 20 programmers and engineers.