❄️Your guide to AI, Holiday Season Special
Hello from London 🇬🇧! I’m Nathan Benaich. Welcome to the Holiday Season Edition covering Q4 2018. I’ll synthesise a narrative that analyses and links important news, data, research and startup activity from the AI world. Grab your beverage of choice ☕ and enjoy the read!
Do hit reply if you’d like to chat about building AI-first products, new research papers, interesting market opportunities or if you’re considering a career move in the startup world.
🎫 As 2019 approaches, make sure to pre-register your interest for RAAIS 2019 using this form (June 28th). We’re lining up a diverse group of entrepreneurs and researchers who are researching and developing the playbook for AI-first technology applications.
🎫 On January 31st 2019, we’re running London.AI #14. The event will feature Alex Caccia, CEO of Animal Dynamics; Philip Torr, CSO of FiveAI and Professor at Oxford; Mostafa ElSayed, CEO of Automata. Register using this form.
Referred by a friend? Sign up here. Help share by giving this it a tweet :)
🆕 Technology news, trends and opinions
🚗 Autonomous everything
AV-focused sensor companies are coming in hot numbers! One of the latest is Aeva, a Bay Area company offering a single rooftop mounted box that packs a velocity-sensing LiDAR and camera operating under 100 watts. The pitch is to help AVs detect the speed of objects in their environment.
Rolls-Royce is one of the latest vehicle manufacturers to enter the self-driving arena, albeit focused on autonomous cargo ships. The company announced a collaboration with Intel to help them deploy by 2025.
Tesla announced it is rolling out an advanced feature as part of Software Version 9.0 called Navigate on Autopilot. The feature can guide a car from a highway’s on-ramp to off-ramp, including suggesting and making lane changes, navigating highway interchanges, and taking exits (all with driver supervision and consent through actions with the turn stalk). Perhaps more impressive than the feature alone is the cabin, software and UX design of the product! Their in-house designed AI chip has yet to be completed.
Several AV companies updated their commercial plans this quarter. Waymo launched their robotaxi service called Waymo One in Phoenix, Arizona. 400 consumers can order a taxi in a geofenced area. Each car has at least one safety driver in it. The prices are also competitive with the ridesharing apps, which suggest the company is spending serious money...Here’s a full customer review.
Volvo is resuming their shipments of XC90 vehicles to Uber’s self-driving unit. Meanwhile, Cruise says they’re on track for a geofenced deployment next year. Cruise convinced Honda to spend $2.75B on their project.
Daimler and Bosch will launch a pilot robotaxi program in San Jose in H2 2019.
In a glimpse forward to how vehicle-to-vehicle communication and fleet data could help reduce traffic incidents, Las Vegas saw a 20% reduction in car crashes as a result of a predictive analytics solution run by Waycare (with Waze data). Waycare’s traffic management solution pulls data on dangerous roads, hazards and incidents generated by Waze users, and augments this with road cameras to manage city-scale traffic.
💪 The giants
Amazon is in office opening mode! The company announced a two new HQs: one in New York City and another in Arlington, Virginia, which together reflect an investment of $5B and more than 50,000 new jobs. Meanwhile, on the other side of the pond, Amazon is growing their Cambridge, UK based Development Centre that is responsible for Amazon Alexa software, AWS, Prime Air and machine learning software. Taken together, Manchester, Edinburgh and Cambridge will grow capacity by 1,000 high-skilled ML-related roles.
🍪 Hardware
As per a recent job ad, Facebook seems to be developing robotics systems within their datacenter infrastructure teams. The role oversees mechanical, electrical, mechatronics, HRI, firmware & software engineers focused on “solving complex problems relating to robotics development initiatives”. This is a noteworthy move displaying how the physical infrastructure powering large-scale internet companies, like that powering e-commerce logistics, needs automation driven by robotics.
ABB, a leading manufacturer and supplier of robotics equipment (eg. robotic arms), are spending $150M to build a robotic-powered manufacturing “factory of the future” in Shanghai. The output of this facility will be, predictably, more robots. The piece notes that in 2017, one in three robots sold in the world went to China (=138k units). ABB and the Shanghai municipal government also signed a strategic cooperation agreement focused on supporting “industry, energy, transport and infrastructure in the region, and to support the “Made in Shanghai” manufacturing initiative”. This suggests a further onshoring of AI-related technology into China. ABB seems to be a committed partner to that end; the company was the first global robot supplier in China to have a complete local value chain including R&D, manufacturing, system integration and service. China is the second largest market for ABB, which has invested $2.4B in the country since 1992.
Graphcore and Dell announced the first joint Dell-Graphcore IPU-Appliance for enterprise data centers. According to the company, the IPU-based platform contains 8 C2 IPU-Processor PCIe cards, each with 2 Colossus GC2 IPU-Processors. The platform delivers over 2 petaflops of AI-specific compute, spread across over 100,000 independent parallel programs, and with a memory bandwidth of nearly 1 Petabyte/sec, for dramatically higher performance and energy efficiency. Watch this short runthrough video of their reference design presented at NeurIPS 2018.
Walmart and SoftBank-funded Brain Corporation agreed to a commercial partnership to deliver 360 commercial floor scrubbers in the US by Q1 2019. BrainOS offers autonomous navigation, data collection and analytics capabilities to the robots. In case you’re interested, another business in this space is called Avidbots.
Results from the new MLPerf hardware benchmark are out and NVIDIA achieved the best performance in the six MLPerf benchmark results it submitted for. Their DGX1 and DGX2 beat Google’s TPUv2.8 and TPUv3.8 (at various scales) and sometimes by a several-fold margin.
🏥 Healthcare
CB Insights profile financing activity of startups using AI-based products to tackle disease diagnosis, drug discovery and development, and clinical trials. It’s worth noting that both 2017 and 2018 were big years for this segment: BenevolentAI raised $120M, Recursion Pharmaceuticals raised $85M and Atomwise raised $54M. There are now well over 50 companies that are active in the market, which is almost 5x the number back in 2015. Moreover, the FDA cleared at least three AI-based diagnostics applications: MaxQ AI’s intracranial haemorrhage device (link), Imagen OsteoDetect for wrist fracture detection (link), and IDx-DR for diabetic retinopathy detection from eye scans (link).
I think we’ll see news of a drug that was discovered or significantly optimised by AI-based methods making progress through very early clinical trials. However, such drugs are not likely to come from big pharma and more likely to come from new entrants. Case in point: only 17 people from big pharma attended NeurIPS 2018! I also think that we’ll see the small-scale deployment of AI-based diagnostic software in 2019, most likely in China first. The biggest outstanding issues will be around a) creating datasets that truly capture human diversity, b) figuring out solid business models that enable independent companies to scale across borders, c) robust clinical trials and validation from lighthouse medical centers. More challenges here.
Researchers at Imperial have used reinforcement learning to extract implicit knowledge from bodies of patient data that far exceeds the amount human clinicians would see in a lifetime. They use this RL agent, the AI Clinician, to learn the optimal treatment plan by analysing many real-life treatment decisions and their outcomes. Interestingly, they show that treatments for sepsis (complications from infection) can be more reliably selected by the AI clinician than human clinicians, thereby reducing fatalities. What’s interesting here is a) AI agents can synthesise experience to much greater extents than any single clinician, and b) AI agents can provide helpful indicators of clinical practice for doctors to more reliably deliver care.
Deep learning methods are making their way into life science research. A group at Brown developed a model called DeepSweep, which is tasked with predicting signatures of natural selection in the genome. Understanding these genomic regions and the traits they endow will help us close the genotype-phenotype gap.
If that’s not impressive enough, deep learning was recently used as a decoding framework for brain-computer interface systems that map brain inputs to discrete actuator controls (e.g. a robotic arm). The subject of this study was a tetraplegic patient with a microelectrode array in the hand and arm area of his left primary motor cortex. The authors were able to show that a neural network decoder could sustain high accuracy reanimation of his paralyzed forearm with functional electrical stimulation sustained for over a year without the need of supervised updating (thus reducing daily setup time).
And if that’s not impressive enough either (!), it turns out that neural networks can predict the 3D structure of proteins from their raw amino acid sequences pretty well. The study (AlphaFold) was submitted to the Critical Assessment of Techniques for Protein Structure Prediction competition where it won first place and produced a doubling of the usual rate of improvement according to this critical analysis piece (it’s worth reading for more insights into the approach and how it compares). As it relates to drug discovery, 3D protein structures are important because they facilitate the modelling of interactions between candidate drugs and their desired target. However, producing 3D protein structures is expensive, cumbersome and slow in part because it requires generating, purifying and running X-ray crystallography and computational modelling to produce. As a result, very few proteins have had their 3D structures characterised. And while we’re at it, DeepMind also published their Science paper on generalising a single AlphaZero model to play shogi, Go, and chess (all in one!).
🇨🇳 AI in China
Taiwan, the powerhouse of semiconductor fabs, is preparing to release details of “millions of Chinese hacks with private companies” to help train AI-based cybersecurity software to predict and prevent future cyber attacks. This is interesting, not least because a) China opposes Taiwan’s international recognition as an independent state and b) Taiwan and China were the allegedly at the source of a major hacking project involving semiconductors that power the servers of major US technology (this report is itself being called into question).
Jonathan Tepperman, chief editor of Foreign Policy, shares concerns about China’s growing power play driven by Xi Jinping. In China’s Great Leap Backward, he points to four factors that are at risk of degrading the country’s incredible economic progress. Given that China plays into almost every AI-related technology discussion today, I think it’s worth considering these factors and how they might curtail the nation’s progress: 1) Continued consolidation of power around Xi Jinping, 2) Ditching the incentive-based bureaucratic system of old with one driven by fear, 3) Discouraging local governments—at the village, county, and provincial levels—to experiment with new initiatives, from building free markets four decades ago to allowing private land ownership more recently, and 4) Limiting the ability for Chinese officials (and even school teachers) to attend foreign meetings and conferences and account for their time abroad on an hour-by-hour basis.
Without intending to create a mountain out of a molehill, I wanted to share one of Jeff Ding’s translations about a recent hire at Pony.ai from Waymo. The piece concerns Zhang Yimeng, who was most recently tech lead for street-level object detection and classification at Waymo and before that Google. In explaining her decision to join Pony.ai, Zhang most strongly considered “the China factor”, which I have anecdotally also seen in person when I travelled to China recently. The translated piece reads: “This girl from the northeast (of China), though she has always been studying abroad and working in Silicon Valley, also hopes to participate further in boosting China’s AI. And she has to admit that, for autonomous cars, China's data, application scenarios and technical challenges are unmatched in the world. Therefore, regardless of whether it was the right time and place, or pursuit of pushing technology further a step, there was no better fit than returning to the motherland.”
Leaning on this sense of purpose and that “now is the time” is important for China’s AI companies. In particular, Chinese AI companies are recruiting from the large diaspora of US-trained AI talent on the West Coast. As noted by James Peng, Pony.ai’s CEO, “Silicon Valley is definitely the place to be...That’s where all the talent is. China has lots of raw talent, but with hardcore AI, it takes years to build up. China has work to do”.
Microsoft Research Asia celebrates 20 years of presence in China. They were early to set up shop and have consequently certainly played a key role in China’s AI, especially on the talent front. According to a translation by Jeff Ding, MSRA has “trained more than 4,800 Chinese interns and more than 500 of them are now active in various large companies in China's IT industry, including Baidu, Tencent, China Mobile, Alibaba, Lenovo, etc. Over 100 people teach at leading universities in China, such as Tsinghua University, Peking University, University of Science and Technology of China, and the Chinese Academy of Sciences.”
Unsurprisingly, the heat behind the China vs. USA semiconductor cold war (not capitalised yet) continues. Adding to this, Canada arrested Huawei’s CFO on accusations that the company broke US sanctions on Iran. The story is evolving.
NYT profile the proliferation of a new type of factory: not one that produces physical products, but one where human annotate data on behalf of companies training machine learning models. “We are the assembly line 10 years ago,” says the owner of a Chinese data labelling company.
AI around the 🌍
Under the UK’s Governments AI Sector Deal, worth up to almost £1B, the Alan Turing Institute received £50M for new Turing AI Fellowships that can attract AI researchers from around the world to the UK. The UK is also establishing a new Centre for Data Ethics and Innovation. Its mission is to identify the measures needed to strengthen and improve the way data and AI are used and regulated. This will include articulating best practice and advising on how we address potential gaps in regulation. Importantly, the Centre “will not, itself, regulate the use of data and AI - its role will be to help ensure that those who govern and regulate the use of data across sectors do so effectively”.
The German government released a plan to invest €3B in AI R&D through 2025 under a plan called “AI made in Germany”. It aims to strengthen research, accelerate commercial translation, create data infrastructure, expand international collaboration and lead the debate on ethics and the impact of AI. Hopefully more details to follow…
California State passed a bill to express the California Assembly’s support for the 23 Asilomar AI Principles as guiding values for the development of artificial intelligence and of related public policy. The State also introduced a bill called the California B.O.T. Act of 2018. This would “criminalize the use of bots to interact with a California person “with the intent to mislead” that person “about its artificial identity for the purpose of knowingly deceiving the person about the content of the communication in order to incentivize a purchase or sale of goods or services in a commercial transaction or to influence a vote in an election.”
🔮 Where AI is heading next
Simulation continues to be a hot topic as a means of improving the performance of AI agents when they’re deployed in the real world. Game engines, which are used to create, render and run game worlds and assets, are throwing their hats into the AI ring. The latest announcement comes from Unity, the world’s most popular gaming engine, which is now working with DeepMind. Unity is growing this application division under Danny Lange, who has hired 80% of his 25 people (or so) ML team in the last two years.
Furthermore, NVIDIA is seeing growing interest in their autonomous driving simulation platform, DRIVE Constellation. Their hardware-in-the-loop setup is intriguing. First, a simulator running on a DGX system generates synthetic sensor data of a car in a virtual world. This data flows into an onboard AV computer system. It is processed and returns driving commands that are sent back in real-time to control the virtual vehicle travelling in the simulated environment. In doing so, developers can validate and test their self-driving stack.
AI safety research is also on the rise and for good reason. This inaugural post by DeepMind’s Safety Research team frames the problems into three categories: 1) Specification, 2) Robustness, and 3) Assurance. I like this problem structure because it mirrors the steps of AI development, testing, deployment and monitoring, while also having clear examples of what happens when AI systems go awry.
In Specification, our major task is to build systems that behave in ways we intended. The downside to getting this wrong is an AI system that might successfully optimise a goal that yields a behaviour we don’t actually want. For example, in a boat racing game, the boat might achieve the highest score by exploiting a bug in the game’s design. A solution framework for this problem, especially as it relates to extremely complex tasks that do not have an obvious training signal to enable data-driven learning, is what OpenAI calls “iterated amplification”. The basic idea is that we could build up a training signal for big tasks from human training signals for small tasks, using the human to coordinate their assembly.
In Robustness, we want our AI systems to be resistant to external manipulation or performance drift. For example, an autonomous car’s vision systems shouldn’t be fooled by adversarial stop signs.
In Assurance, we need adequate levels of diagnostic power and control over our AI systems. For example, it makes sense to have off switch if all other methods of intervention fail us.
Numenta released a new theory of intelligence: The Thousand Brains Theory of Intelligence, in which they posit that the neocortex does not learn a single world model but instead learns complete models of individual objects and concepts in the world. The authors suggest that “cortical grid cells” allow the neocortex to learn these object models that operate in a distributed, parallel fashion. This contrasts with deep learning-based theories of the brain, in which the environment is perceived through a series of layered processing steps that extra increasingly higher level features.
Gary Marus expands upon a recent tweetstorm: “It worries me, greatly, when a field dwells largely or exclusively on the strengths of the latest discoveries, without publicly acknowledging possible weaknesses that have actually been well-documented.” I think that’s fair, but also doesn’t really need to escalate to religious warfare on Twitter :) I also believe that François Chollet’s talk at RAAIS 2018 on proposes an elegant solution framework.
🔬 Research
Here’s a selection of impactful work that caught my eye:
Optimizing Agent Behavior over Long Time Scales by Transporting Value, DeepMind. How often have you reflected on the outcome a decision that you made a week or a month ago? Learning from our successes and mistakes requires linking actions and consequences over long spans of time (the “credit assignment problem”). Doing so is key to our ability to learn efficiently. In this paper, the authors explore this feature of human nature that is otherwise absent in AI models today that can only reason over short timescales. The authors introduce a new paradigm for reinforcement learning that is based on the following three principles where agents must: 1) Encode and store perceptual and event memories; 2) Predict future rewards by identifying and accessing memories of those past events; 3) Re-evaluate these past events based on their contribution to future reward. Their system, Temporal Value Transport (TVT), integrates these requirements by using neural network attentional memory mechanisms to credit distant past actions for future rewards. According to the authors, “the algorithm is not without heuristic elements, but we prove its effectiveness for a set of tasks requiring long-term temporal credit assignment over delay periods that pose enormous difficulties to conventional deep RL.”
Visualizing and understanding generative adversarial networks, ICLR 2019 submission. The authors sought to identify an interpretable structure to GANs that could provide a window into their internal representations. Interestingly, they found many parts of GAN representations that can be interpreted to have causal effects on the synthesis of objects in the output image. Of note, these interpretable effects can be used to compare, debug, modify, and reason about a GAN model. This could make its way into photo editing software, for example.
Large-scale GAN training for high fidelity natural image synthesis, ICLR 2019 submission. Some seriously pretty class-conditional samples generated synthetically by their BigGAN. And some more here from NVIDIA Research.
Deep Imitative Models for Flexible Inference, Planning, and Control, Berkeley. The authors present Imitative Models, a class of probabilistic predictive models that can plan expert-like trajectories to achieve arbitrary goals. They combine the flexibility benefits from model-based RL (MBRL) and the ability of imitation learning to learn from human demonstrations. They show that deep imitative models substantially outperform both direct imitation and classical MBRL in simulated driving tasks and can be learned efficiently from a fixed set of expert demonstrations.
ChauffeurNet: Learning to Drive by Imitating the Best and Synthesizing the Worst, Waymo Research (blog post here). In this study, the authors seek to build a system to drive by imitating an expert. They created a deep RNN, ChauffeurNet, that is trained to generate a driving trajectory by observing a “mid-level representation of the scene” as an input. Interestingly, this representation is not built directly from raw sensor data. As such, it factors out the perception task and allows Waymo to combine real and simulated data for easier transfer learning.
Measuring the Effects of Data Parallelism on Neural Network Training, Google Brain. Continuing my search for empirical evidence around data requirements to build high-performance ML models, I bring you this paper. Here, the authors explore the exact relationship between the batch size (e.g. how many images to use for each pass through a neural network) and how many training steps are necessary. They explore how this relationship varies with the training algorithm, model, and data set. The study finds that for idealized data parallel hardware, there is a universal relationship between training time and batch size, but there is dramatic variation in how well different workloads can make use of larger batch sizes. On the similar topic, OpenAI publish results on how gradient noise predicts the parallelizability of neural network training on a wide range of tasks.
Beyond imitation: Zero-shot task transfer on robots by learning concepts as cognitive programs, Vicarious AI. This paper learns concepts as programs on a 'visual cognitive computer', showing zero-shot task-transfer on robots, and strong generalization. Vicarious bring ideas from cognitive science about perceptual symbol systems and image schemas into the realm of machine learning.
Episodic Curiosity through Reachability, DeepMind, GoogleAI, ETH Zurich. In this paper, the authors propose a novel episodic memory-based model of granting rewards to reinforcement learning agents when they are solving tasks that otherwise have sparse rewards. The rewards in this model are akin to curiosity, i.e. encouraging the RL agent to explore the environment but also solve the original task. They do this by enabling the RL agent to store observations from its interactions with the environment in memory. The reward is then calculated as a function of how far the current observation is from the most similar observation in memory. Specifically, a neural network is trained to predict whether two observations were experienced close together in time or far apart. If the agent makes observations not already stored in memory, it gets more reward.
Exploration by Random Network Distillation, OpenAI and Edinburgh. Inspired by related challenges than the Episodic Curiosity paper above, this study introduces an “exploration bonus”, where we predict the output of a fixed randomly initialised neural network based on the current observations of a deep RL agent in a game environment. This bonus is based on the notion that neural networks tend to have significantly lower prediction errors on examples similar to those on which they have been trained. This motivates the use of prediction errors of networks trained on the agent’s past experience to quantify the novelty of the new experience. The authors reach state of the art performance on Montezuma’s Revenue:
Model-Based Active Exploration, NNAISENSE. A third paper released in the same week on RL methods that encourage exploration to improve agent learning!
Reward learning from human preferences and demonstrations in Atari, OpenAI and DeepMind. This paper seeks to address how agents can solve tasks where the subgoals are poorly defined or hard to specify as a hard-coded reward function. This is important because most complex tasks in life don’t have an obvious reward function that can be written down. They solve this by initializing the agent’s policy with imitation learning from the expert demonstrations using the pretraining part of the DQfD algorithm (Hester et al., 2018). Second, they use trajectory preferences and expert demonstrations to train a reward model that lets them improve on the policy learned from imitation.
Three reviews of EMNLP 2018, a leading natural language processing conference: here (Patrick Lewis, UCL), here (Sebastian Ruder, Aylien) and here (Claudia Hauff, TU Delft).
Here’s a list of the best paper awards at NeurIPS 2018.
What can a machine infer from a single image? First, it was depth estimation (the team now works at Niantic Labs). Now, it’s 3D object detection (this team works with Wayve.ai).
Visual Model-Based Reinforcement Learning as a Path towards Generalist Robots: This paper presents an algorithm that can learn a general-purpose predictive model using unlabeled sensory experiences. They show that this single model can enable a robot to perform a wide range of tasks. Sergey Levine’s group also released their SOTA off-policy model-free reinforcement learning algorithm, soft actor-critic (SAC). It is arguably one of the most efficient model-free algorithms available today to enable robotic locomotion, object manipulation and more.
📑 Resources
Blog posts
Bringing machine learning research to product commercialisation: a worthwhile resource from Rasmus Rothe at Merantix. The playbook on this topic continues to be written!
Owkin published a neat primer for physicians to understand the fundamentals of ML and an example use case of predicting the age of a patient’s brain from MRI scans.
Libby Kinsey published a neat overview of the challenges of infrastructure plumbing for AI-first applications and current solution providers that are on the market. While we’re still in early days of seeing AI-first products in the wild, current deployments are already pointing to key issues with existing infrastructure that was developed for the traditional web SaaS/consumer/marketplaces application environment. In particular, this excerpt from a Google paper illustrates some differences to how we should think about testing and monitoring:
Lyft’s Level 5 team authored a nice introduction to the constituents of HD maps and what they’re used for.
What are the challenges with analysing biological and artificial neural networks? Read more here!
NYU’s AI Now Institute published their latest 2018 report. The study focuses on the “accountability gap” in AI between creators and users, AI for surveillance and government’s use of automated decision systems, fairness, bias and discrimination, as well as unregulated AI experimental on human populations. The authors identify emerging challenges within these topics and provide practical solution pathways informed by research so that policymakers, the public, and technologists can better understand and mitigate risks.
A new Pew survey of nearly 1,000 tech experts found that fewer than two-thirds expect AI technology to make most people better off in 2030 than today. Many express a fundamental concern that AI will specifically be harmful.
Robot Economy: Ready or Not, Here It Comes: A technological framework describing a robot economy is outlined and the challenges it might represent in the current socioeconomic scenario are discussed.
Videos/lectures
DeepMind’s lectures at UCL in London on “Advanced Deep Learning and Reinforcement Learning” are live on YouTube.
Datasets
VoxCeleb is the largest audio-visual dataset of human interviews from YouTube. It includes over 7,000 identities, over 1 million utterances and over 2,000 hours of video with audio files, face detections, tracks and speaker meta-data available. It’ll be useful for speaker recognition in video, speech separation, face synthesis and emotion recognition.
Researchers at the Allen Institute for AI in Washington have contributed a dataset to further visual common sense reasoning. The dataset includes 290k multiple choice QA problems derived from 110k movie scenes. In particular, the authors present an approach for transforming rich annotations into multiple choice questions with minimal bias.
A practical approach to learning machine learning, all in notebooks!
Open source tools
A series of Jupyter Notebooks providing a step-by-step introduction to data science and machine learning.
While the code behind the majority of DeepMind’s key programs continue to be closed source, the company’s research engineering team released a new TensorFlow-based library of building blocks for writing RL agents.
Two new libraries for private machine learning in TensorFlow (paper) or PyTorch (link).
Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models (link). This is a new effort for AI-driven drug discovery led by Insilico Medicine.
OpenAI released CoinRun, a new RL training environment that provides a metric for an agent’s ability to transfer its experience to novel situations.
💰 Venture capital financings and exits
Q4 2018 has been an active one! Here’s a highlight of the most intriguing financing rounds:
Zymergen, a Bay Area company with a mission to search beyond the bounds of human intuition to deliver novel products and materials, closed a $400M Series C investment led by SoftBank’s Vision Fund (SoftBank proper led their Series B). This business is often held up as a posterchild for how the software industry can intersect with life sciences to accelerate what is an otherwise slow discovery and development process. The key idea is to use machine learning (and other computational methods with lab automation) with real-world and simulated data to navigate a vast search space of potential (in)organic molecules to make discoveries in a much more directed fashion. Zymergen now has over 500 employees, 42% of whom are in research and 21% are in engineering.
Horizon Robotics, a Chinese semiconductor company focused on embedded computer vision processors for large-scale facial recognition in security and mobility, is on the market to raise $1B at $3B-$4B valuation. Horizon is yet another Chinese AI startup founded by former members of Baidu’s self-driving car project (another example being PonyAI).
Automation Anywhere, one of the big 3 robotic process automation vendors, raised a $300M extension to their $250M Series A that was announced only in July. This round is led by SoftBank Vision Fund at $2.6B post-money while the prior one was led by Goldman Sachs and NEA at $1.8B post-money.
DataRobot, vendor of a predictive analytics platform for enterprises to build and deploy predictive models the cloud or on-premise, continues its rapid growth by raising a $100M Series D led by Sapphire Ventures and Meritech.
A few semiconductor related financings:
Esperanto Technologies, a maker of energy-efficient RISC-V computing solutions, raised a $63M Series B;
Habana Labs, the Israeli maker of a separate training and inference AI processors, raised a $75M Series B led by Intel along with Bessemer and Battery Ventures.
Wave Computing, which provides dataflow-based systems to “eliminate the need for a host and co-processor in the processing of a neural network”, raised $86M in Series E financing.
Two businesses raised significant capital to apply AI techniques in the sales arena. First, Afiniti raised $130M at a $1.6B valuation. On the one hand their product analyses sales people’s performance with specific types of calls and situations, and on the other hand, it analyses customers’ prior interactions with a company. It then matches up customer service reps who it believes will be most compatible with specific customers. Second, Chorus.ai raised a $33M Series B from Georgian Partners to help sales teams have “higher quality conversations that result in higher quota attainment, higher rep productivity, and shorter new hire ramp time”.
Apex.ai, led by Jan Becker (a longtime Bosch AV engineer and member of Sebastian Thrun’s Darpa robotics challenge team), raised a $15.5M Series A led by Canaan and Lightspeed to build an operating system for self-driving cars.
AEye, whose iDAR sensor couples a solid state LiDAR and high-resolution camera in a single device to detect and track moving vehicles up to a kilometer away (!), raised a $40M Series B led by Taiwan’s government-backed venture fund.
Standard Cognition, an Amazon Go-style autonomous checkout system (yeah, it’s becoming Uber for X at this point), raised a $40M Series A led by Initialized Capital.
Primer, which automates the understanding of large text corpuses in the enterprise, raised $40M Series B led by Lux Capital to expand into new verticals outside of fintech and government.
Appzen, an AI-based solution for auditing employee expenses in the enterprise back office, close a $35M Series B led by Lightspeed.
💰A couple of M&A deals, including:
Cylance, a Bay Area cybersecurity company focused on predicting and preventing the execution of advanced threats against endpoints, was acquired for $1.4B by Blackberry in the latter's bid to evolved into a software-only vendor. Cylance’s technology is deployed over 14.5 million endpoints in including many Fortune 100 organizations and governments. The company was founded in 2012 by Ryan Permeh (formerly Chief Scientist at McAfee) and Stuart McClure (formerly EVP and Worldwide CTO at McAfee). The business has almost 1,000 employees and raised $297M.
Blue Vision Labs, a London-based post-Series A stage company building city-scale 3D maps for shared AR experiences, was acquired by Lyft’s Level 5 (self-driving) team. The deal is worth $72M with an extra $30M on the basis of milestones. BVL built a team of 40 people, many of whom have significant computer vision, robotics and machine learning experience from Oxford and Imperial, as well as large tech companies. The company had opened their developer SDK in Q1 2018 ahead of Niantic’s Real World Platform that was announced in June 2018.
Silk Labs, a US startup focused on on-device AI to empower businesses to build the next generation of intelligent connected devices, was acquired by Apple for an undisclosed amount. The company was founded by two former Mozilla alums and a Qualcomm alum in 2015. The team ran a Kickstarter campaign where they raised $150k (after $4M in VC capital) for an AI-enabled smart camera and hub, called Sense. The product’s USP was that all the processing was done locally on a Qualcomm chip. The product never shipped and money was returned to backers.
---
Signing off,
Nathan Benaich, 16 December 2018
Air Street Capital | Twitter | LinkedIn | RAAIS | London.AI