Attribution methods assess the contribution of inputs to the model prediction. One way to do so is erasure: a subset of inputs is considered irrelevant if it can be removed without affecting the prediction. Though conceptually simple, erasure's objective is intractable and approximate search remains expensive with modern deep NLP models. Erasure is also susceptible to the hindsight bias: the fact that an input can be dropped does not mean that the model `knows' it can be dropped. The resulting pruning is over-aggressive and does not reflect how the model arrives at the prediction. To deal with these challenges, we introduce Differentiable Masking. DiffMask learns to mask-out subsets of the input while maintaining differentiability. The decision to include or disregard an input token is made with a simple model based on intermediate hidden layers of the analyzed model. First, this makes the approach efficient because we predict rather than search. Second, as with probing classifiers, this reveals what the network `knows' at the corresponding layers. This lets us not only plot attribution heatmaps but also analyze how decisions are formed across network layers. We use DiffMask to study BERT models on sentiment classification and question answering.

Members get 60 days free site access, $6.95/article thereafter. Airbnb is colliding with traditional hotel companies like Marriott International and Hilton. In just over a decade, the online lodging marketplace has assembled an inventory of more than 7 million rooms -- six times as much lodging capacity as Marriott managed to accumulate over 60-plus years. In terms of U.S. consumer spending, Airbnb overtook Hilton in 2018 and is on track to move ahead of Marriott.1 Although Airbnb serves similar consumer needs, it is a completely different kind of company.

Video-based interviews, either on Skype or Facetime or as recorded links, have been used for some years, but now a new and tricky feature has become popular: HR tech companies have begun to deploy artificial intelligence (AI) to assess candidates even as their skills are being tested on several parameters. Companies are using AI-based voice recognition and facial recognition software that reads the candidate in a way that no human can. When a candidate answers questions, AI software reads expressions and measures voice modulation to get into the mind of the candidate. Lying and misleading expressions can be caught. Candidates hear an automated voice and see no one at the other end as they get time to answer questions while looking into the camera.

Sometimes HR managers and recruiters have a tough time finding the right candidate for a job. Can conversational AI save the day? Recruiting the right talent is one of the most challenging tasks for recruiters. On top of these responsibilities, recruiters have to develop a recruitment strategy that meets business goals, takes into consideration competitor analysis and employee satisfaction rates. In reality, managing this workload effectively is a hard task to do, that's why many companies turned to Artificial Intelligence for extra help. In one of their recent reports, Deloitte covered a use case of a multinational bank struggling to optimize the workload of hundreds of service desk agents, who support HR management processes.

Marriott International, the world's largest hotel chain, announced earlier this month a rollout of facial recognition software for check-ins in partnership with Chinese tech giant Alibaba Group. The new kiosks at two Marriott locations in China will be able to scan and identify guests' faces, pull up their reservations, and check them in--all without the help of humans.

China is spearheading the use of facial recognition for everything from helping control major live events to ordering fast-food, but also bolstering a growing domestic surveillance system that has raised fears among human rights activists of privacy being invaded.

Alexa has a new job: hotel concierge. Amazon has launched a version of Alexa for hotels that lets guests order room service, ask for more towels or get restaurant recommendations without having to pick up the phone and call the front desk. Marriott has signed up for the service, and will place Amazon Echo smart speakers in 10 hotels this summer. It is another way for Amazon to sell its voice assistant to businesses and get Alexa in front of customers. Amazon already sells a version of Alexa for workplaces, and has struck deals to place Alexa in cars and refrigerators.

Alexa has a new job: hotel concierge. Amazon has launched a version of Alexa for hotels that lets guests order room service through the voice assistant, ask for more towels or get restaurant recommendations without having to pick up the phone and call the front desk. Marriott has signed up for the service, and will place Amazon Echo smart speakers in 10 hotels this summer, including its Westin and St. Regis brands. It is another way for Amazon to sell its voice assistant and devices to businesses and get Alexa in front of more customers. Amazon already sells a version of Alexa for workplaces, and has struck deals to place Alexa in cars and refrigerators.

Alexa has a new job: hotel concierge. Amazon has launched a version of Alexa for hotels that lets guests order room service through the voice assistant, ask for more towels or get restaurant recommendations without having to pick up the phone and call the front desk. Marriott has signed up for the service, and will place Amazon Echo smart speakers in 10 hotels this summer, including its Westin and St. Regis brands. It is another way for Amazon to sell its voice assistant and devices to businesses and get Alexa in front of more customers. Several media reports had said that Marriott had tested both Apple Inc's Siri and Amazon's Alexa to select what was best suited for its hotels.

Getty Images has announced a partnership with AI platform Cortex, to create a solution for creatives that recommends the best Getty Images content for social media campaigns. The program launched last month and has allowed creatives to better scour the database's 80m still images and illustrations. This has been possible through the AI's analysis, which has already allowed over 33,000 brands to predict what color schemes, imagery, and keywords perform best on social media. Cortex currently has a list of customers across several industries that includes Heineken, Hulu, Jack Daniels and Marriott. Results have come in from the launch already: Cortex has said in a statement that marketers using the tool to create content outperformed their industry benchmarks by up to 300% and save an average of 8.5 hours per user per week.