Scientists from the University of Technology Sydney (UTS), developed a highly-sensitive nano-thermometer that uses atom-like inclusions in diamond nanoparticles to accurately measure temperature at the nanoscale.
Yes, I know. I’ve been gone for a month. Very eventful month it is. In the meantime, I’ve dumped all the links to interesting news read in April. Quick one liners:
The new “allelic drive,” is equipped with a guide RNA (gRNA) that directs the CRISPR system to cut undesired variants of a gene and replace it with a preferred version of the gene. The new drive extends scientists’ ability to modify populations of organisms with precision editing. Using word processing as an analogy, CRISPR-based gene drives allow scientists to edit sentences of genetic information, while the new allelic drive offers letter-by-letter editing.
Deep Science AI merges with Defendry. Defendry was working on establishing automated workflows for internet of things devices including notifying a company when triggered by dangerous situations like weapons, fires, and robberies, or situations suggesting that a crime is about to begin. Similarly, Deep Science AI’s computer vision system could spot guns, masks, and other dangerous triggers from a CCTV footage.
IBM develops ‘zero-noise extrapolation’ – repeating computation at varying levels of noise to estimate what a quantum computer would calculate in the absence of noise. More accurate and complex computations can be achieved without any hardware improvements. Increasing circuit depth (a measure of the number of sequential operations performed on the processor) leads to increased errors due to decoherence (loss of quantum coherence / information from a system in the environment). Zero-noise extrapolation technique mitigates the effect of decoherence while benefiting from increased circuit depth.
Reinforcement learning requires a large amount of interactions in simulated or real-world conditions for learning to occur. Google proposes a new algorithm – Simulated Policy Learning (SimPLe) – using game models to learn quality policies in selecting actions. SimPLe learns a world model of how the game behaves and uses this model to optimize a policy for action taking (via model-free reinforcement learning). SimPLe uses medium-length rollouts to minimize prediction error. It takes four frames as input to predict the next frame along with the reward. After training, sample sequences of actions, observations, and outcomes are used to improve policies. The model-based reinforcement learning methods mitigates high cost and slow human labeling.
Apple’s new credit card promises privacy. Goldman Sachs will use customers’ personal data to operate the card, but the data will not be shared or sold to third parties for advertising. Apple claims they would not know where customers have shopped, what they have paid, or bought. Spend tracking and categorization happens on-device, and not Apple servers – supposedly.
Companies are reluctant to shift their systems on Oracle databases to competitors like Amazon’s AWS or Microsoft cloud services. Concerns include fear of Oracle finding software license contract violations. Penalty with fees or terminating permission to use software may ensue. Software asset management is a branch of consulting that helps companies negotiate software contracts.
In Hong Kong, wealth is accumulated in property-owning magnates, unsurprisingly. The government, incentivised by revenue from land and property tax, have little reason to move into a new phase of development for the land-starved others. I have always felt strongly for Hong Kong citizens living in cage-like quarters.
WeWork has more than doubled its sales to $1.82 billion in 2018. Its losses also doubled to $1.93 bilion, with $46 million of its expenses for “transaction fees and inventory write-off”including fees for bankers and the SoftBank transactionm as well as inventory for building materials.
DARTS (Deep-learning Augmented RNA-seq analysis of Transcript Splicing) uses deep-learning based predictions on public datasets of RNA sequencing to provide insights into alternative splicing. Variations in alternative splicing may cause or exacerbate diseases. Although massively parallel RNA sequencing is now a standard technology used to investigate alternative splicing, it would require a large amount of sequences and high cost for accuracy.
Scientists in US and Russia have managed to return three qubits to their state a fraction of a second earlier, demonstrating time reversal in a quantum computer. Qubits were set to a state of 0, after which order is lost and qubits grew in complexity in a pattern of 0s and 1s. Then, time was reserved by a program that allows the QC to go from chaos to order (complex to simple). Qubits, in this way, were sent back in time. Success rate was 85% with two qubits, and falls with every qubit added, indicating that this time-reversal is negligible in the material world. However, time reversing has bearings on quantum computer programs – while we could not measure its state without destroying superposition and projecting a component with some probability, we have a method of time-reversing the QC to its initial state with this demonstration.
Through computer simulations, scientists show that we could possibly travel through certain types of black holes – specifically large, rotating ones – without being killed. Hyperspace travel could be possible within our milky way or even between galaxies.
Zeus is the AirBnb for white collar transient workers, raking in 45 millon in revenue and raising $24 million in their latest round. Its managed ownership model means that Zeus takes over furnishing and upkeep while homeowners profit. It fits into the niche area between AirBnb’s short term stay and other larger, long-term apartment rentals.
This issue is a couple days in making! I haven’t come across much news that intrigued me.
Appen acquires Figure Eight, for a total of up to $300 million. Both companies use crowdsouced labor to annotate data which is then used to in training AI and machine learning models.
StyleGAN aims to generate images that are indistinguishable from real ones using GAN. Rather detailed technical piece that I will revisit in-depth!
Silot, a Singapore-based fintech startup, closes $8m in their Series A. Silo organizes banks’ previously siloed data. It feeds the data, including policies, into a decision engine. For Krungsri Bank, Silot’s pilot test helped detect transaction fraud and money laundering, and automated onboarding and verification processes.
One way I describe it to my girls is, ironically, by taking your ego out of what you do, it will liberate you and you can ultimately perform better and more comfortably.
Some really great articles this morning! Quantity and quality. But – before that, there has always been a photo close to my heart and it made it to the (e-)papers today!
On this set is Charles Mingus, Thelonious Monk, Charlie Parker – in my opinion the Big Three besides Everyone’s Favorite John Coltrane of course. The three have found a spot in my heart because of how dynamic and non-conformist, even for jazz standards, they are. Monk’s discordance (check out Body & Soul, you’ll see what I mean 10 seconds in), Parker’s unfettered energy (listen to Salt Peanuts! Salt Peanuts!), Mingus’ buttery smooth but mischievous chords. I haven’t much exposure to Roy Haynes – also in the set. Sorry, Roy. Yes, this is absolutely the greatest photo in jazz history
Ethics in Technology
Collated findings from the Moral Machine Experiment – a global survey on who or what we should spare over others in the event of a self-driving car crisis. Results are not surprising, with humans – and vulnerable humans – at overwhelming advantage of being spared. The variations across culture are in the degree of preference to spare pets, passengers / pedestrians, more people over less, etc. All pretty consistent with prior findings of cultural preferences and mindset. But worth a think too: who would you save over another, and why?
Liao, professor of bioethics at NYU, proposes the possibility of genetic engineering humans to reduce our carbon footprint (although he apparently does not endorse it) in his paper “Designing humans: A human rights approach“. Suggestions include engineering aversion to redmeat, reducing human size, enhancing altruism. He also introduces principles to lay ground for this possible future. I quite like this quote:
They’re right to be concerned,” he says. But in a world where such technology exists, he asks, “do we want a society where we say, ‘Nobody can have it’?
Liao seems careful in expressing a clear stand on this contentious topic, but he seems at least semi-open to this future. Frankly, I don’t feel the instinctive disgust towards genetic engineering as ‘unnatural’ like many others do. I do tend towards the futuristic, and think that it’s an inevitability. What I’m most concerned about is regulation, fair play, and possible cases of abuse.
Biotechnology
Researchers at Osaka University developed a DNA sequencing method that could distinguish the drug molecules from normal nucleotides in short strands of DNA. In what they call the ‘single-molecule quantum sequencing’ method, they could identify individual molecules in the DNA based on electrical conductance variations. What are the implications of detecting anticancer drug molecules incorporated in the DNA? Ability to understand the molecular damage done to DNA by anticancer drugs, and from there develop more effective drugs.
I have so much admiration for researchers and doctors who advance biomedical therapies! It’s a costly and long-drawn affair, sometimes without headway even as they fight for funding. But the output and the goal is amazing – until some corporation or another snaps it up for exploitative profit-making, I mean. 🙃
The Stowers Institute for Medical Research uses big data analysis on proteomics data to better model protein interaction networks. A topological scoring (TopS) algorithm, combines data sets to identify proteins that meaningfully interact. In lay man terms, there is such a huge amount of protein that detecting that rare interaction that is important in producing an effect is challenging. Now, researchers can better understand how biological processes are carried out. The TopS algorithm can also be applied to previous biological data generated – or even other fields of biotech / science!
Neuroscience
Neuroscientists are able to predict a choice up to 11 seconds before the decision is consciously made. The study suggests that ‘traces of thoughts’ that sit unconsciously in our brain can influence us to make a choice that is reinforced by the stronger ‘trace’, with implications on PTSD, uncontrollable thoughts, unwelcome visualizations in OCD.
Studies like such are not new – there have been findings on how we can predict motor decisions and actions up to 10 seconds before the conscious decision. Usually, they are framed as having implications on our free will, which I disagree with. As above, I chose to focus on the distinction between unconscious and conscious thoughts, and their interaction. Free will is hardly relevant to this discussion – not even, as the article points out, our conception of free will. I’m fairly certain everyone is aware we have unconscious and subconscious thoughts and that they influence our decisions. The fact that we have a choice – no matter the degree – is free will, isn’t it?
Great article to kickstart questions on anthropomorphism. The emotions animals are capable of and how that gives us greater insight on what being human is, what differentiates us? Also, how do we represent the inner worlds, mental models, and ‘consciousness’ of nonhuman animals? In neuroscience we know that the determining factor lies in the developmental phase of the prefrontal cortex and our vs their ability for advanced introspection and self-awareness. I’m of the belief that animals have consciousness, although that depends on your definition of consciousness.
Participants were made to observe prehistoric ceramics from different cultural and societal styles, while measuring their visual behavior (eye movement) as an indication of cognitive processing. Findings indicate significant parallels between the visuals of different cultures and the cognitive processing styles. For instance, the shape and form of objects and the pattern of visual exploration has a clear evolution through time, and may be connected to social behavior and complexity.
I love evolutionary psychology and this particular study speaks to me. It’s amazing how we can use historical artefacts to now extrapolate how cognitive processes have evolved over time (through visual behavior).
Chimpanzees have their own unique set of local traditions and cultures, but these are slowly eroded with human intervention and observation. In the short amount of years we have studied animals through intervention, we have taken away the diversity of their behavior. Ed Yong writes a great number of articles on this topic which are equally fascinating. I’ve always been interested in anthropological studies through extrapolation from our primate counterparts.
A compilation of tips for product managers towards successful integration. This is so relevant to my previous job, where deployments happened monthly. Integrations start with learning the API of the system, building and testing codes for the integration, and performing regression tests everytime the vendor updates. Of the tips, I found: prioritizing long-term roadmaps, documentations, and supporting users’/customer’s big-picture plans and goals to be the most pertinent. Although they are not particularly ‘agile’ (have to plan long-term and document extensively), I feel the pain of non-sustainable integrations that require continuously patching and bug catching. Will relook this if and when I go back to product management.
Applications of AI in customer experience is becoming the new standard. This article breaks down the ways in which AI could be utilized. Nothing new to me, since they are what I do at work every day! But worth noting for anyone new to tech applications in customer experience. Data unification and analyses allows extraction of customer insights. These enable tailored recommendations based on purchase behavior; personalized customer support (chatbots, virtual assistants) and user experience based on individual customer behavior and contextual factors. Beyond that, the article mentions anticipating customer obstacles and automating routine processes for greater efficiency. Personally I have not worked on the latter, but it is worth digging into.
Transposit raises $12.2 million in Series A. It is an API composition platform that serves as a universal translation layer for developers to compose their APIs. It uses relational databases, with a Java backend translating SQL into optimized execution (whatever that means). It’s also able to handle pagination, authentication, and retries in the background.
Research on image generation AI models found a solution to bypass ground-truth labels. It first uses a semantic extractor to extract feature representations, then perform cluster analysis on the data. Training is done with a GAN (two part neural network consisting of generators and discriminators) to infer labels than rely on hand annotated labels. This is paired with co-training – combining unsupervised, self-supervised, and semi-supervised methods. We’re moving towards unsupervised AI and away from data tagging and labeling reliant AI training – seen here too.
A quick one today because I’ve been busy, and I haven’t found much news that sparked my interest beyond that of data privacy.
DuckDuckGo is an encrypted search engine that doesn’t store your personal information or track you. It also has voice search features and displays search results for no-click answers. Its growth (9 billion searches in 2018) indicates a shift in user focus on privacy.
Mark Zuckerberg’s apparent pivot of Facebook into privacy is interesting. This article neatly summarizes its implications: a potential focus on commerce for revenue instead? While this article points out that Facebook responds to user desires, in this case — privacy.
Google Ad Manager prepares to simplify programmatic ad management. It will be transiting to a unified first price auction under Google Ad Manager, by the end of 2019. It claims to be a step towards more transparent market, an advertising buyer’s bid will not be shared with another buyer before the auction or be able to set the price for another buyer.
This upcoming week I hope to write a bit about a Design Thinking Academy I attended, and maybe a little more about agile organizations and workstreams!
GPipe, Google’s open source library to train deep neural networks, aims to increase classification accuracy and training speed. It uses two training techniques: synchronous stochastic gradient descent, an optimization algorithm that updates an AI model’s parameters, and pipeline parallelism, a task execution system where one step’s output is streamed as input to the next step.
Customer data, particularly of current and predicted customer preferences, are polled to be the most desired data by companies. Challenges, however, include integrity and reliability of data.
Wefox, a Berlin insurtech company, allows independent insurance brokers to onboard clients onto the platform, automating the administrative processes. It also provides a D2C insurance platform, allowing modular, timestamped customization of insurance packages for direct customers. Consider looking into PolicyPal and GoBear and what their VP offerings are.
Somewhere between Airbnb and apartment rental, Blueground allows shorter and more flexible leases. Consider viability and existing similar services in Singapore.
Q.bit 