Finally, feature-level alignment is performed between the aggregated domain and target domain while training the segmentation network.

Simulation-to-real domain adaptation for semantic segmentation has been actively studied for various applications such as autonomous driving. Existing methods mainly focus on a single-source setting, which cannot easily handle a more practical scenario of multiple sources with different distributions. In this paper, we propose to investigate multi-source domain adaptation for semantic segmentation. Specifically, we design a novel framework, termed Multi-source Adversarial Domain Aggregation Network (MADAN), which can be trained in an end-to-end manner. First, we generate an adapted domain for each source with dynamic semantic consistency while aligning at the pixel-level cycle-consistently towards the target. Second, we propose sub-domain aggregation discriminator and cross-domain cycle discriminator to make different adapted domains more closely aggregated. Finally, feature-level alignment is performed between the aggregated domain and target domain while training the segmentation network. Extensive experiments from synthetic GTA and SYNTHIA to real Cityscapes and BDDS datasets demonstrate that the proposed MADAN model outperforms state-of-the-art approaches. Our source code is released at: https://github.com/Luodian/MADAN.

Neural Information Processing Systems http://nips.cc/

We sincerely thank all the reviewers for their insightful comments to help us improve the paper. T o Reviewer #2. 1. Are multiple sources more beneficial? This is largely due to the fact that domain gap also exists among different source domains. We will reorganize the layout of Figure 1 in the main paper to make it more clear. We thank the reviewer for pointing this out.

Simulation-to-real domain adaptation for semantic segmentation has been actively studied for various applications such as autonomous driving. Existing methods mainly focus on a single-source setting, which cannot easily handle a more practical scenario of multiple sources with different distributions. In this paper, we propose to investigate multi-source domain adaptation for semantic segmentation. Specifically, we design a novel framework, termed Multi-source Adversarial Domain Aggregation Network (MADAN), which can be trained in an end-to-end manner. First, we generate an adapted domain for each source with dynamic semantic consistency while aligning at the pixel-level cycle-consistently towards the target.

Nightfall AI, a startup providing cloud data loss prevention services, today announced that it raised $40 million in Series B financing from investors including WestBridge Capital, Venrock, Bain Capital Ventures and -- for some reason -- athletes and celebrities including Paul Rudd, Drew Brees and Josh Childress. CEO Isaac Madan says that the proceeds will be put toward doubling Nightfall's 60-person headcount, scaling the platform to more customers and markets, and expanding Nightfall's partner ecosystem. Isaac was previously a VC investor at Venrock, where he focused on early-stage investments in software as a service, security and machine learning. Rohan was one of the founding engineers at Uber Eats, where he designed and built software to grow the platform's footprint. Madan says he and Sathe were inspired to launch Nightfall by Sathe's personal experiences with data breaches arising from poor "data security hygiene."

Simulation-to-real domain adaptation for semantic segmentation has been actively studied for various applications such as autonomous driving. Existing methods mainly focus on a single-source setting, which cannot easily handle a more practical scenario of multiple sources with different distributions. In this paper, we propose to investigate multi-source domain adaptation for semantic segmentation. Specifically, we design a novel framework, termed Multi-source Adversarial Domain Aggregation Network (MADAN), which can be trained in an end-to-end manner. First, we generate an adapted domain for each source with dynamic semantic consistency while aligning at the pixel-level cycle-consistently towards the target.

While some facets of the sci-fi classic have come to pass, such as voice-command technology, and others are currently unfolding, such as the destructive effects of climate change, the dramatic AI advances that fool Harrison Ford still feel a long way off. But that hasn't stopped AI from becoming one of the buzziest terms thrown around by technology and financial services experts. Fintech startups use the word to make products appear cutting-edge and established firms find they have to offer "AI" to remain competitive. The products they are referring to as AI, though, are often little more than new data models or even just outright vaporware -- technology sold on a conceptual level that hasn't been built yet, according to many technology leaders in the advice industry. "There are a lot of packages out there that are more'artificial' than'intelligence'," said Raj Madan, BNY Mellon Pershing's managing director of technology.

Nightfall (formerly Watchtower AI), a San Francisco-based provider of cloud data loss prevention solutions, today emerged from stealth with $20.3 million in funding led by Bain Capital Ventures and Venrock, with participation from Pear VC and Atlassian CTO Sri Viswanath. CEO Isaac Madan said the proceeds from this latest round will bolster Nightfall's R&D and market expansion. "Our mission at Nightfall is to build the control plane for cloud data, enabling enterprises to know what data they have across the cloud, and to proactively manage and protect that information," said Madan, who was previously with Venrock's investment division. Nightfall's eponymous platform monitors data flowing into and out of the services a customer uses, which machine learning algorithms classify as sensitive, personally identifiable (PII), noncompliant (with regulations like HIPAA and GDPR), or safe to share. Using a visual dashboard, admins can set up automated workflows for quarantines, deletions, alerts, and more that integrate with dozens of software-as-a-service (SaaS) platforms and APIs, and view analytics metrics like real-time and historical PII count by type.

For those suffering depression, PTSD, or other mental health challenges, a breakdown can be a slow burn, developing over days or weeks in between doctor visits. Delay in treatment can have lasting consequences. "Very likely, every single episode of depression or mania is going to damage your brain a little more," says Thilo Deckersbach, a Harvard psychology professor who practices at Massachusetts General Hospital. The hospital's online MoodNetwork recruits patients with major depression and bipolar disorder for clinical studies, including a new one with the Boston-based artificial intelligence company Cogito. They are testing the ability of a mobile app called "Companion" to flag early signs of trouble by monitoring activities like patients' movement, calling and texting behavior, and the way they speak.