Google open-sources AI that searches tables to answer natural language questions

Google today open-sourced a machine learning model that can point to answers to natural language questions (for example, “Which wrestler had the most number of reigns?”) in spreadsheets and databases. The model’s creators claim it’s capable of finding even answers spread across cells or that might require aggregating multiple cells.

Much of the world’s information is stored in the form of tables, points out Google Research’s Thomas Müller in a blog post, like global financial statistics and sports results. But these tables often lack an intuitive way to sift through them — a problem Google’s AI model aims to fix.

To process questions like “Average time as champion for top 2 wrestlers?,” the model jointly encodes the question as well as the table content row by row. It leverages a Transformer-based BERT architecture — an architecture that’s both bidirectional (allowing it to access content from past and future directions) and unsupervised (meaning it can ingest data that’s neither classified or labeled) — extended along with numerical representations called embeddings to encode the table structure.

A key addition was the embeddings used to encode the structured input, according to Müller. Learned embeddings for the column index, the row index, and one special rank index indicate to the model the order of elements in numerical columns.

Google AI table searchGoogle AI table search

Above: A table and questions with the expected answers. Answers can be selected (#1, #4) or computed (#2, #3).

Image Credit: Google

VB Transform 2020 Online – Live July 15-17, 2020: Join leading AI executives at VentureBeat’s AI event of the year.
Register today and save 30% off digital access passes.

For each table cell, the model generates a score indicating the probability that the cell will be part of the answer. In addition, it outputs an operation (e.g., “AVERAGE,” “SUM,” or “COUNT”) indicating which operation (if any) must be applied to produce the final answer.

To pre-train the model, the researchers extracted 6.2 million table-text pairs from English Wikipedia, which served as a training data set. During pre-training, with relatively high accuracy, the model learned to restore words in both tables and text that had been removed — 71.4% of items were restored correctly for tables unseen during training.

After pre-training, Müller and team fine-tuned the model via weak supervision, using limited sources to provide signals for labeling the training data. They report that the best model outperformed the state-of-the-art for the Sequential Answering Dataset, a Microsoft-created benchmark for exploring the task of answering questions on tables, by 12 points. It also bested the previous top model on Stanford’s WikiTableQuestions, which contains questions and tables sourced from Wikipedia.

“The weak supervision scenario is beneficial because it allows for non-experts to provide the data needed to train the model and takes less time than strong supervision,” said Müller.

Typical Entertainment’s Thunder Rally is pure multiplayer mayhem

Presented by Intel

It’s a tough time for a lot of people around the world right now, and the creators of Thunder Rally hope that their multiplayer game can provide some relief to friends and families stuck at home.

Out now on PC via Steam Early Access, Thunder Rally is a free-for-all vehicular brawl in which up to eight players try to obliterate each other using cars and ridiculous power-ups. It’s been a labor of love for independent studio Typical Entertainment –  the developers have been making Thunder Rally in their spare nights and weekends for over two years.

The team is a mix of current and former employees from The Void, the maker of location-based virtual reality games like Avengers: Damage Control and Star Wars: Secrets of the Empire. But while they enjoy being at The Void, Typical Entertainment’s founders were itching to make a game of their own, one inspired by their work in game jams and frequent sessions of Super Smash Bros. 

VB Transform 2020 Online – Live July 15-17, 2020: Join leading AI executives at VentureBeat’s AI event of the year.
Register today and save 30% off digital access passes.

“We were all playing Super Smash Bros. — it was before the newest one came out — and it wasn’t up to eight players. We always had game nights. … The fifth guy was always left out and we were like, ‘You know what we should do? We should make a game that we can all play all at once,’” said Typical co-founder Abel Perry.

The developers took inspiration from other games as well. Co-founder Eric Davis drew on his time playing custom demolition derby-style multiplayer maps in Halo, where players would ram each other off an arena while driving Warthogs, and those who died could still compete by tossing grenades from the sidelines. Typical wanted Thunder Rally to have that same kind of non-stop experience.

Thunder Rally PC Game

Thunder Rally PC Game

“Everyone has a chance to keep playing with basically no downtime,” said Davis.

Fighting til the bitter end

While each round of Thunder Rally starts off as a simple demo derby, it doesn’t stay that way for long. In addition to colliding with one other, you can drive over crates to pick up power-ups that give you new offensive or defensive options. Some are just weapons like landmines and homing missiles. But others, like the rewind time ability (resets your car back to a previous position) and jets that allow your car to briefly fly, can drastically alter your strategies.

So-called “super” power-ups, such as a black hole and a Transformers-like mech mode, can turn the tide of battle in your favor. The longer a round lasts, the more frequent these super power-ups will appear — it’s Typical’s way of keeping matches short and snappy. And if that wasn’t enough, dead players return as catapults that spawn on the edges of the map, and they can launch explosive barrels at anyone who’s still alive.

Some of the levels also have environmental hazards that you must avoid on top of dodging your fellow players (or CPU-controlled cars if you don’t have enough people). With that many variables, a match can quickly devolve into pure pandemonium.

Thunder Rally PC Game

Thunder Rally PC Game

“We want Thunder Rally to be pick-up-and-play, but we also want it to have this bit of difficulty so that if you’re really good, you can just smash everyone. But it still has enough randomness to give low-skill players a chance to win,” said Perry.

Growing with Early Access

Balancing day jobs with indie development hasn’t been easy for the team. But Typical’s partnership with Intel means it doesn’t have to shoulder that burden alone. The developers have been using Intel’s quality assurance services, and the company also helped them make sure that Thunder Rally runs well on integrated GPUs like the UHD Graphics 630 chip.

That frees up Typical to focus on Thunder Rally’s Early Access period. With 40 levels and several car designs, the game already has a lot of content. But the team still has big updates in the works, all of which will expand Thunder Rally way beyond their original aspirations.

“Thunder Rally was supposed to help us learn the ropes of the process and just get something out there as quickly as possible. But it gained a bit more traction than we thought it would,” explained Davis. “And so we decided, ‘Okay, let’s go all the way with this. Let’s try to get this out to as many consoles as we can. Let’s try to get online multiplayer.’”

Getting online multiplayer up and running is one of the studio’s major priorities (currently, you can only play with friends through Steam’s Remote Play Together feature). Typical is also using Early Access to collect feedback from the Thunder Rally community and see what changes or improvements need to be made.

“Even though there is just one mode, there are a lot of levels and power-ups and we continue to keep adding more so that the game feels fresh and you can play it for a long time,” said Davis. “We’ve been playing it for two years, and we still have a blast every time we play.”

Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. Content produced by our editorial team is never influenced by advertisers or sponsors in any way. For more information, contact

Vida Health raises $25 million to treat chronic health conditions remotely

Vida Health, a virtual health care platform that connects people with chronic ailments to professional therapists and coaches, has raised $25 million to meet the “increased demand” for remote care — particularly around mental health.

As with other virtual health-focused platforms, Vida said that it has seen a spike in demand for its service during the COVID-19 pandemic, which has pushed millions of businesses and consumers into lockdown. Indeed, the company said that since COVID-19 was declared a pandemic in early March, it has launched to more than 500,000 new end users, bringing the total number of people covered by its platform to over 1.4 million.

Founded out of San Francisco in 2014, Vida targets those with conditions such as diabetes, depression, anxiety, and high blood pressure, and pairs them with a personal trainer, nutritionist, nurse, therapist, or other relevant support professional. The company’s platform also leverages machine learning to crunch data and personalize each user’s treatment plan. The Vida app can integrate with many popular health and fitness apps and devices, including Fitbit and Apple Watch, giving coaches direct access to their users’ vital stats — with this data, they can develop and tailor plans as the data changes over time. Built-in messaging and chat functionality also allows them to communicate in real time.

Enterprises represent a major part of Vida’s target market, and it’s typically offered as part of a company’s health care benefits. Vida also sells device “bundles” that include contraptions such as heart-rate monitors, scales, blood pressure cuffs, activity trackers, glucometers, and more.

Above: Vida Health bundle

Health boom

VB Transform 2020 Online – Live July 15-17, 2020: Join leading AI executives at VentureBeat’s AI event of the year.
Register today and save 30% off digital access passes.

Prior to now, Vida had raised $53 million, and with another $25 million in the bank it’s well-financed to capitalize on what will likely prove to be a boom period for the virtual health care industry. Pre-pandemic figures from MarketsAndMarkets had the global telehealth market pegged as a $25 billion industry, and this had been expected to double within five years. However, with the coronavirus forcing industries across the spectrum to adapt to social distancing, the telehealth market could be set to grow exponentially.

In the past few weeks alone, we’ve seen some sizable funding rounds plowed into the remote health care realm. Tyto Care, for example, raised $50 million to grow its telehealth examination and diagnostic platform, which includes a health kit that anyone can use at home to test their lungs, heart, throat, ears, skin, abdomen, and body temperature and transmit the data to their doctor. And earlier this week, Medici, which is a little like WhatsApp for remote medical care, secured $24 million.

Vida’s latest funding round was led by global life science investment firm Ally Bridge Group (ABG), with participation from Yahoo cofounder Jerry Yang’s AME Cloud Ventures, Nokia-backed NGP Capital, Aspect Ventures, Canvas Ventures, Webb Investment Network, and Workday Ventures.

Sign up for
Funding Weekly to start your week with VB’s top funding stories.

Gaming sees explosive growth in social video as people stay home

Entertainment, in all of its forms, has always been an important pastime for Americans. And now with millions sheltering at home, it’s taken an even more crucial role in helping people escape the boredom and anxiety that comes with a global pandemic. As a result, there have been huge increases in not only TV consumption but also gaming. And it’s not just about playing solo: Social video has given rise to replays, promotions, live-streamed speedruns, and more that connect everyday players to the gaming franchises and influencers they love most.

A new report from social video analytics company Tubular Labs shows just how much social video growth for gaming has occurred since people began sheltering in place. First, some topline stats.

During the week of March 30, gaming content on YouTube had its best week ever — with 17 billion views, representing a 24% year-over-year growth. Facebook has also seen an increase in gaming videos, although not as much as YouTube.

When it comes to the performance of various types of gaming content, there’s growth across the board, with the highest jump (35%) belonging to online games. (In the chart below, performance is based on Tubular’s V7 rating, which is the average number of views per video within the first seven days of upload, comparing pre-and post-COVID-19.)

VB Transform 2020 Online – Live July 15-17, 2020: Join leading AI executives at VentureBeat’s AI event of the year.
Register today and save 30% off digital access passes.

Game-specific trends

Tubular also examined trends around some of the most popular games: Call of Duty, Minecraft, and Fortnite.

There have been 60% more Call of Duty videos uploaded on YouTube since people began sheltering in place in March, particularly around its Warzone’s Season Three. The pandemic has also given new players a chance to rise in the ranks: Four of the top five creators of Call of Duty content from March 10 through April 10 were not in the top five in the previous 30-day period.

Of course, Call of Duty: Warzone launched March 10, so in addition to benefiting from folks staying home, the community is also excited for this new battle royale experience.

Minecraft has seen more modest growth (6% in videos uploaded), but one creator in particular, Skeppy, has seen viewership skyrocket by an impressive 285%.

Independent creators aside, official brand channels are an important part of the puzzle. And from Tubular’s analysis, it’s clear that Fortnite is revving up: The official channel has uploaded twice as many videos during the pandemic than it did in the 30 days prior, putting it in the top five for overall Fortnite creators. (Note: This was compiled before the Travis Scott concert.)

Views for sponsored content rise during pandemic

As marketers across myriad industries scrambled to revamp advertising strategies amid COVID-19, gaming is one area in which they’re latching on. Overall, 249 brands have worked with 457 gaming partners since the pandemic began, and per video performance is up, with the top campaigns delivering more than 1 million average V7 views. Looking at YouTube specifically, the week of March 30 was the most-viewed week for sponsored gaming content this year, and the sixth most-viewed week in the last three years.

Viewership of esports surges

With traditional live sports all cancelled, the pandemic could be esports’ time to shine. The week of April 6 saw the most views on esports content since the week of the Fortnite World Cup Finals last year. Live esports content draws viewership from an engaged crowd, those live videos are making an impact on demand as well. Tubular found that live esports videos that are published later generate a 20% higher average V30 rating and are 40% more engaging.

When it comes to watching live streams, Twitch is one of the most popular platforms, and it’s only gaining steam in the pandemic. In the weeks after the outbreak, live streaming of gaming creators reached up to 869 million minutes watched per day across all viewers, a 20% growth from two weeks prepandemic. Who and what are they watching?

The Electronic Sports League leads for minutes watched, followed by creators summit1g and loltyler1; highest watch time by game goes to Call of Duty: Modern Warfare and League of Legends.