Deepmind Develops an AI System That Finds a Way Around Simulated Cities It Hasn’t Seen Before

Venturebeat | June 17, 2020

  • DeepMind says it designed a system that can leverage prior knowledge to solve tasks.

  • A grand challenge in AI is architecting a model that’s able to enter unfamiliar environments and get to work immediately.

  • EPN leverages self-attention, a method for computing relationships among an arbitrary number of items that doesn’t assume any particular structure among them.

DeepMind says it designed a system that can leverage prior knowledge to solve tasks, while at the same time exploring to gather new knowledge and plan using this new knowledge when faced with new tasks. In a paper accepted to the Conference on Computer Vision and Pattern Recognition (CVPR) 2020, researchers at the company describe an AI “planning module that operates over episodic memories (memories of everyday events that can be explicitly stated), which they say outperforms the nearest baseline by two to three times with respect to planning and exploring.

A grand challenge in AI is architecting a model that’s able to enter unfamiliar environments and get to work immediately. For example, the paragon household robot would use general knowledge about homes to find cleaning supplies and acquire information it anticipates will be useful, like the location of clothes hampers in the rooms it passes. It could then leverage the newfound knowledge (i.e., hamper locations) to plan solutions for future tasks (e.g., doing the laundry) that solve the tasks more quickly.

Unfortunately, even state-of-the-art episodic memory models are able to explore but not to plan, potentially because they lack mechanisms for planning using memories. DeepMind claims to have remedied this with a novel module — episodic planning network (EPN) — that prompts AI agents to explore and plan effectively in unfamiliar environments.


EPN leverages self-attention, a method for computing relationships among an arbitrary number of items that doesn’t assume any particular structure among them. EPN begins with episodic memories that reflect experience in a scenario so far, with each memory containing representations of the current observation, the previous action, and the previous observation.

In an experiment that brings to mind the New York City-navigating AI that Facebook open-sourced two years ago, the DeepMind researchers trained EPN-based software agents in One-Shot StreetLearn, a simulation where environments are sampled as neighborhoods from Google’s StreetLearn data set of real-world street-level imagery. In One-Shot StreetLearn, you define tasks by selecting a position and orientation that the agent must navigate to from its current position.

Given only an image showing the current location, an image representing the goal location, and the ability to move left, right, or forward, the EPN-based agents successfully reached 28.7 goals per episode (averaged over 100 consecutive episodes) in places unfamiliar to them, according to the coauthors. They also matched the minimum number of steps to complete new tasks after only 15-20 tasks, and they generalized well to larger neighborhoods containing a greater number of intersections, with performance reaching 77% success with nine intersections as opposed to five in the original tasks.

In the current experiments, the agent could succeed by planning over observed states,” the researchers wrote. “However, there is nothing preventing EPNs from being used to plan over belief states, a potential critical ability for operating in dynamic partially-observed environments … Future work might may approach [problems] with broader task distributions … and test the extent to which EPNs are effective in solving broader classes of tasks.

EPN builds on DeepMind’s existing city-navigation work and Dreamer, which internalizes a world model and plans ahead to select actions by “imagining” their long-term outcomes. More recently, the lab detailed Agent57, a system that uses episodic memory to learn a family of policies for exploring and exploiting. (Agent57 is one of the first systems to outperform humans on all 57 Atari games in the Arcade Learning Environment data set.)



The Year Neural Networks Were Born Warren McCulloch and Walter Pitts proposed the model of artificial neurons. The Name ‘Artificial Intelligence’ Came into Existence American computer scientist John McCarthy first adopted the name at the Dartmouth Conference.

Other News

Cogito Announces Updates to Emotion AI Software on Salesforce AppExchange, the World's Leading Enterprise Cloud Marketplace

Cogito | June 15, 2022

Cogito, the leader in real-time coaching and guidance for the enterprise, today announced it has updated its Emotion AI software on Salesforce AppExchange, providing customers new ways to add real-time coaching and guidance cues to the Salesforce Lightning Experience, and to incorporate personalized coaching plans into Service Cloud Workforce Engagement. This integration creates a unified desktop, giving agents more freedom to navigate customer calls without the distraction of multiple tools and systems. Cogito's Emotion AI software extracts over 200 signals from every call, applying AI models that generate real-time cues. The in-the-moment guidance helps agents handle complex interactions, generate personalized coaching recommendations and provides supervisors visibility into agents working from anywhere, including alerts on frustrated callers or burned out representatives. Available on Salesforce AppExchange since 2017, the latest integration provides users a way to generate more informed and more cohesive call experiences. The latest enhancements are designed to further arm agents with real-time support and allow supervisors to manage the workforce more effectively, improving the well-being of employees and customer experiences. “The modern contact center needs more real-time support and insights to enable better conversations, better work experiences and better outcomes at enterprise scale. “We’ve brought together some of the best minds in behavioral science, machine learning, and enterprise technology to create an AI solution that provides real-time emotional intelligence, and offers the right interventions to lift entire customer-facing organizations up. I’m thrilled our team can now offer the enhanced product capabilities seamlessly through the Salesforce Service Cloud integration to better support organizations navigating customer interactions.” Josh Feast, CEO and co-founder of Cogito “We are excited that Cogito is continuing to innovate on AppExchange as they extend the scope of agent well-being to the tools and systems supporting the user experience. Agents should have coaching guidance and customer data in one place to improve the employee experience in tandem with the customer experience," said Woodson Martin, GM of Salesforce AppExchange. "AppExchange is constantly evolving to meet the needs of our customers, and we love watching our partners evolve alongside us." About Salesforce AppExchange Salesforce AppExchange, the world’s leading enterprise cloud marketplace, empowers companies, developers and entrepreneurs to build, market and grow in entirely new ways. With more than 7,000 listings, 10 million customer installs and 117,000 peer reviews, AppExchange connects customers of all sizes and across industries to ready-to-install or customizable apps and Salesforce-certified consultants to solve any business challenge. About Cogito Cogito innovates emotion AI and conversation AI, combining them to provide contact center teams and other frontline customer support professionals insights into not only what is being said but how things are said in phone conversations. Cogito features powerful behavioral and lexical models that provide real-time coaching and guidance to call center agents and sellers, gives front-line supervisors visibility into live conversations of their teams working from anywhere, and continuously monitors customer sentiment. Cogito is used by the largest contact centers including 8 of the Fortune 25, to improve customer conversations and employee well-being. Founded in 2007, Cogito is a venture-backed software company located in Boston, MA.

Read More


Headroom launches to combat Zoom fatigue with AI

Headroom | December 16, 2021

To say that the global pandemic has transformed the workforce would be something of an understatement, with remote work still a reality for millions of people globally. To help with the rapid transition from centralized working environments to a more distributed network of home-working hubs, cloud-based software-as-a-service (SaaS) tools such as Zoom have shown their worth. But at what cost? In a world where major businesses, from Salesforce and Microsoft to VMware and Dropbox, have confirmed a permanent shift to a remote-first or hybrid working policy, people are spending less time face-to-face and more time face-to-screen. As such, better and more adaptive virtual collaboration tools will prove to be vital — Zoom fatigue, after all, is very real. That is precisely what Headroom is setting out to solve, with an AI-powered platform that automates many of the laborious tasks involved in virtual video meetings, and removes some of the stress entailed in catching up with key discussion points. Headroom was formed in 2020 by a triumvirate of founders with experience in building products at companies such as Google and Magic Leap. CEO Julian Green previously founded an AI-powered photo startup called Jetpac, which Google acquired in 2014. He then went on to become a group product manager at Google, where he launched the Google Cloud Vision API and Android Vision on-device API, as well as overseeing various augmented reality visual search products. More recently, Green served as a director at Alphabet’s “moonshot factory” X. Headroom raised $5 million in seed funding last October, with big-name backers, including Google’s AI-focused fund Gradient Ventures and Yahoo cofounder Jerry Yang. As of this week, Headroom is officially available for anyone to sign up for after an extended beta period. Zoom on steroids While Headroom is a video communication platform in a similar vein to Zoom, it packs a number of notable tools on top of its core video chat functionality — it has been built with productivity, inclusiveness, and organization at its core. Headroom automatically transcribes a meeting, enabling users to search the transcript for keywords that are relevant to them — rather than having to sit through an entire two hour replay to get to the stuff that matter’s for their role. With this transcription in place, Headroom can also conduct a textual analysis using natural language processing (NLP), and by leveraging computer vision to analyze body language — including gaze-tracking and facial expressions — to generate highlights based on the perceived energy levels of the participants. Moreover, built-in gesture recognition allows meeting participants to quickly show their (dis)approval without unmuting their microphone, either by giving a thumbs up/down, or raising their hand to indicate that they have a question without verbally interrupting the current speaker. On a technical level, Headroom uses what is known as “real-time super-resolution,” which essentially transforms low-definition video into high-definition using lower bandwidth. And with its server-based architecture, Headroom doesn’t consume users’ local computing power — everything takes place in the cloud. “We can provide all these AI features without taking up all your CPU cycles, and having your computer fan running when you have a meeting,” Green explained. “Our differentiation from the many other efforts in this area stem from having developed a next-generation video-conferencing platform that enables low latency real-time AI to be run on real-time communication — older architectures can’t do this. Incumbents are in the pixel delivery business, we are in the intelligence delivery business.” Meetings of the future In the future, Headroom sees itself serving as a sort of centralized knowledge base that allows users to ask questions — such as “what was project XYZ’s outcome?” — and receive answers. This means those that couldn’t attend a meeting can easily discover any agreements or conclusions that were reached. “Meetings will become a more flexible way to drive a business forward, from rich in-person communication all the way to rapid information-retrieval and value-added analytics. -Green “We have had inbound interest from consultancies, banks, healthcare, and educational companies, So there is value there for enterprises, which are very interested in the things we do well — coaching, improving inclusion, and transparency in meetings. Having a searchable store of meeting information, reducing duplicated meetings, and improving efficiency is attractive to companies of all sizes.” -Green

Read More


Absolute Software Announces New Partnerships with Leading ISVs Leveraging Absolute Application Persistence-as-a-Service

Absolute Software | June 07, 2022

Absolute Software™ , a leader in self-healing endpoint and secure access solutions, today announced new partnerships with leading Independent Software Vendors (ISVs) to strengthen resiliency of their endpoint applications with Absolute Application Persistence-as-a-Service (APaaS). As APaaS partners, Ericom, Utopic, and WinMagic® are leveraging Absolute’s firmware-embedded technology to monitor and automatically heal their mission-critical security solutions - ensuring they remain healthy, installed, and working effectively across their customer bases. Through APaaS, ISVs have the ability to embed Absolute’s Application Persistence™ capabilities directly into their software packages to continuously maintain the integrity of their endpoint applications with lower development costs and no ongoing code maintenance. APaaS partners also have access to rich application health telemetry data that offers valuable insights into application health and performance, allowing for continuous optimization and improvement. “As APaaS partners, Ericom, Utopic, and WinMagic have taken the critical steps needed to harden and self-heal their mission-critical applications at the firmware level. “Our unique device intelligence consistently shows that the complexity of today’s endpoint environments has left endpoint agents under constant threat of decaying, colliding with other applications, or being disabled by malicious or negligent users. By ensuring their endpoint agents remain healthy and working effectively, these leading ISVs are delivering a higher return on software investments and strengthening security posture for their end customers.” Edward Choi, SVP of Global Alliances at Absolute Software “Hardening our ZTEdge platform has been a top priority for Ericom, and with Absolute, we’ve found the right partnership to put our vision into motion,” said John Peterson, CPO at Ericom. “Adding Absolute’s undeletable firmware-based self-healing to our mission-critical endpoint application gives us the confidence that we are delivering on our promise of enabling our customers to strengthen their security defenses in today’s perimeter-less digital environments.” “Our customers depend on our leading self-healing automation capabilities to enforce policies, ensure compliance issues, improve asset performance, and reduce operational support costs as part of their NIST Cybersecurity Response and Recovery journey,” said Bob Whirley, CEO at Utopic. “By partnering with Absolute to harden our endpoint agent with APaaS, we have even more confidence that we are enabling seamless workplace transformation and modernization.” “Some of the world's largest organizations rely on SecureDoc everyday to protect their sensitive data from falling into the wrong hands,” said Thi Nguyen-Huu, CEO at WinMagic. “To complement our encryption offering, we are thrilled to add authentication to our comprehensive security product suite with MagicEndpoint. The ability to leverage Absolute’s firmware-embedded, self-healing capabilities to our leading authentication and encryption suite through APaaS gives us even more assurance that we are delivering on our promise to offer security solutions that free our customers to think, share and achieve their goals, knowing their employees and data are protected.” About Absolute Software Absolute Software is a leading provider of self-healing endpoint and secure access solutions, delivering truly resilient zero trust security for today’s distributed workforces. Absolute is the only endpoint platform embedded in more than half a billion devices, offering a permanent digital connection that intelligently and dynamically applies visibility, control and self-healing capabilities to endpoints, applications, and network connections - enabling customers to strengthen cyber resilience against the escalating threat of ransomware and malicious attacks. Trusted by nearly 16,000 customers, G2 recognized Absolute as a leader in Zero Trust Networking and Endpoint Management in the Winter of 2022.

Read More


Xerox Acquires Canadian IT Services Provider, Powerland, to Grow Presence in IT Services Market

Xerox | February 04, 2022

Xerox Holdings Corporation (NASDAQ: XRX) today announced it has acquired Powerland, a leading IT services provider in Canada, for cash. The acquisition strengthens Xerox’s IT services offerings in North America, which include cloud, cyber security, end user computing and managed services. Powerland serves customers throughout Canada and provides a wide array of technology solutions and services including cloud, hybrid cloud, cyber security, and end-user computing solutions based on individual business needs. Powerland recently earned 2021 industry accolades including Nvidia Partner of the Year, Dell Marketing Partner of the Year North America, HPE Service Partner of the Year and Tech Data Circle of Excellence. Today, the ability to quickly adapt to change and remain competitive requires a strong IT partner, Combining the talented teams of Powerland and Xerox IT Services provides a greater value proposition for customers with support, information security expertise and technology solutions from major IT manufacturers. We manage our customers’ technology so they can manage their business.” Martin Bachant, president, Xerox Canada. Recent research shows that SMBs will increase their investments in key IT solutions such as cloud infrastructure, managed IT services, collaboration solutions and cyber security. The addition of Powerland will enable Xerox to expand reach in North America to offer increased value to customers throughout Canada. We are thrilled to join a world-class innovator like Xerox and look forward to enhancing business outcomes and technology transformations for our clients across Canada.” Curtis Dery, executive vice president, Powerland. This acquisition continues Xerox’s commitment to expand its IT services business. Xerox acquired both ITEC Connect, a leader in IT services in the UK, and Digitex, an IT services provider in Canada, in March 2020. Xerox acquired Competitive Computing (C2), a Vermont based IT services company, in October 2021. The global IT services market is sized at more than $680 billion annually according to Techaisle. About Xerox Holdings Corporation (NASDAQ: XRX) For more than 100 years, Xerox has continually redefined the workplace experience. Harnessing our leadership position in office and production print technology, we’ve expanded into software and services to power today’s workforce. From the office to industrial environments, our differentiated business solutions and financial services are designed to make every day work better for clients — no matter where that work is being done. Today, Xerox scientists and engineers are continuing our legacy of innovation with disruptive technologies in digital transformation, augmented reality, robotic process automation, additive manufacturing, Industrial Internet of Things and cleantech.

Read More


The Year Neural Networks Were Born Warren McCulloch and Walter Pitts proposed the model of artificial neurons. The Name ‘Artificial Intelligence’ Came into Existence American computer scientist John McCarthy first adopted the name at the Dartmouth Conference.