Flora Salim

Hi Kanban Singapore folks, There's an upcoming workshop on Systems Thinking—a powerful approach to understanding and managing complex systems. Systems Thinking helps us see the big picture by recognising patterns and interconnections within systems. This is particularly relevant to Kanban, especially with the STATIK (Systems Thinking Approach to Implementing Kanban) method, which we use to optimise our workflows. By applying Systems Thinking, we can better identify bottlenecks, understand feedback loops, and make more informed decisions to improve our processes. The workshop will cover: - Introduction to Systems Thinking - How to apply it in various contexts - Hands-on practice with causal loop diagrams It's a fantastic chance to deepen our knowledge and enhance our Kanban practices. If you're interested in joining, register via Telegram here: https://lnkd.in/gYydvTiS. Cheers, Debbie #KanbanSingapore #KanbanSG #Kanban #KanbanUniversity #YesWeKanban #LeSS #lessworks

11

1 Comment

Artificial Intelligence (AI) is advancing rapidly, and despite its numerous advantages, the associated risks are creating a pressing need for effective regulation. Proximity’s multi-disciplinary experts understand designing, developing, and reviewing complex and innovative regulatory frameworks. In this article on Consultancy Australia Proximity’s Declan Norrie and Kyle Wood discuss the importance of continuous improvement and collaboration across government portfolios as a necessary step in addressing AI’s safety standards and risk management. Read the full article here: https://ow.ly/xQew30sFwV2

4

(cont. from https://lnkd.in/gRfY-jRt) SPIRE paper: Online Computation of String Net Frequency with Peaker Guo (University of Melbourne), Tony Wirth (University of Sydney, University of Melbourne), Justin Zobel (University of Melbourne) The net frequency (NF) of a string in a text is the number of occurrences of the string in the text with unique left and right extensions. NF was originally introduced for Chinese NLP tasks but there is a lack of understanding of the properties of NF and the absence of efficient algorithms for computing NF. At CPM 2024, we showed that NF is combinatorially interesting and how NF can be computed efficiently in the offline setting. For many applications, however, facilitating the computation in an online manner is highly desirable. We are the first to solve the problem in the online setting, and we do so in optimal time.

6

#Graph databases can uncover intricate relationships hidden in #data, but the key lies in having a graph model that accurately represents the use case. Kelsen Diogenes, during his Master's degree at the University of Auckland - Data Science, developed a property graph #model. This model specifically targets #fraud patterns within publicly available Company Office Data in New Zealand. By encoding suspicious patterns as queries in #Cypher, and executing them against a #Neo4j database, potential instances of fraudulent behavior like beneficial ownership, income tax evasion, money laundering, or Phoenix companies can be identified. Please find a teaser presentation below, and I will send the full report (anonymised) on request: https://lnkd.in/gurDqjWZ

33

1 Comment

There are a few good reasons for Australia to develop our own LLM, including the need to maintain our own culture, and control over the flow of our own information. The primary argument, however, is that LLMs are the infrastructure of an AI-enabled economy, and if Australia is going to make the transition we need to own and control our own infrastructure. The transition is already here, it's just that it's not currently Australian, it's American. The idea that an American should tell us that we should just leave it to them is no great surprise I suppose. It's the modern equivalent of colonialism, and they want to own the railroads.

65

10 Comments

Hi everyone! I’d like to share one of my latest publications on performing Bayesian inference using ABC (Approximate Bayesian Computation), coauthored with David Frazier from Monash University. In this work, we introduce a new method for diagnosing model misspecification in cases where inference is done using ABC. Unlike previous methods, ours is based on the asymptotic properties of ABC. We show, both theoretically and through examples, that our method can reliably detect model misspecification and performs well with finite sample sizes, often outperforming existing approaches. Linnk: https://lnkd.in/e8d-YvKv

63

1 Comment

The government has just announced funding for a National Quantum network ($18.5M). https://lnkd.in/gk-rF27Y. This is great news. It is crucial that the government fund networks of emerging industries since they lack the financial resources to support themselves. But as a roboticist I feel a bit forgotten. I'm singing in my head "What about me?" This announcement came after: - Investment in quantum computing ($1B) https://lnkd.in/gxpXPjUm - Solar Sunshot Manufacturing ($1B) https://lnkd.in/gS3Xnzva - Green Steel Manufacturing ($200M) https://lnkd.in/gVqwU9yW - Acquire Radar Company ($345M) https://lnkd.in/geTHubA4 - National Critical Minerals R&D Hub ($50M) https://lnkd.in/gxce3uxy - National AI Centre in 2021 ($47M) https://lnkd.in/gQ_4xRYX - National Space Agency in 2017 ($2B) https://lnkd.in/g8w7fW5Q And yet, there is no funding for a National RAS (Robotics and Autonomous Systems) Centre, even though RAS has been listed as a Critical Technology for Australia https://lnkd.in/gcMCptKG and one of the Enabling Capabilities for the NRF https://lnkd.in/gP6Wxyb8 And most recently: Future Made-in-Australia https://lnkd.in/gGfe6Ryr Where the PM believed that the "problem of labour costs is diminishing — thanks to robotics." But this ignores the fact that Australia has one of the lowest adoption rates of robotics in the western world (35th) https://lnkd.in/gDGgqRvk Which means if we want to make stuff in Australia, we need to dramatically increase that adoption and deployment of RAS? How is this going to happen? Yes we have some excellent existing initiatives, like the Robotics Australia Group and ARM Hub (Advanced Robotics for Manufacturing) but as far as I know there is no direct federal funding. It is now a year since the consultation on the National Robotics Strategy: https://lnkd.in/gZwbwtPJ" Where it was estimated that “Automation technologies, including robotics, provide an opportunity to add between $170 billion to $600 billion per year to Australia’s GDP by 2030" But if RAS is so critical to Australia - to defence, to mining, to agriculture, to logistics, and to manufacturing, why doesn't the Federal Government fund a National RAS Centre - just like all the other critical technologies. #whatAboutRobotics #futuremadeinAustralia #knowledgeDiffusion #nationalRoboticsStrategy #nationalRASCentre Ed Husic MP Here is a link to related posts: https://lnkd.in/gUj9a-Gq https://lnkd.in/gRenZgMq https://lnkd.in/g2EpBseH https://lnkd.in/gwY_NqJs

21

7 Comments

Amid the news of the Australian Government's ambitious $1 billion investment in quantum computing, my comments were published in The Sydney Morning Herald. The article discussed the issue of quantum threats and the Q-day, You may find it here: https://lnkd.in/gHCb6QZU A part of it, “When there’s a reasonably sized quantum computer that will be able to break a standard encryption mechanism, that will be Q-Day,” Dr Fida Hasan, a UNSW lecturer in cybersecurity, says. Unlike Y2K, we don’t have a time and date for when Q-Day will strike. Some predict it will come within 10 to 15 years, Hasan says. Others reckon it will arrive much quicker as governments pour funds into the quantum computing race, just as we saw on Tuesday with a $1 billion quantum boost for Australia, courtesy of the federal and Queensland governments. #quantum #cybersecurity #postquantumcryptography #pqc #quantumcomputer #australia #1binvestment

21

Did you know two core programming languages for developing AI (R and Weka) were developed in New Zealand? With the announcement of “Apple Intelligence”, as well as Google’s shifts into LLM augmented search, we’ve reached a tipping point for integration of generative AI. A timely report by the Artificial Intelligence Researchers Association is probably the best thing I've read on national AI strategy for New Zealand, exploring how to maximise benefits and minimise risks for AI adoption, including driving economic opportunities and productivity in New Zealand. Building off Apple’s announcement and the AIRA report, here are some key trends I’m seeing in both. 🔍 Informed Consumer Choices: Privacy-conscious consumers will seek products that minimise AI integration, or demand greater control and transparency. 🏛️ Governance and Values: Companies will emphasise the governance and ethical features of their products. Apple’s announcement that its Privacy Cloud product includes independent audits of privacy and security exemplify this trend, as does its use of on-device processing and use of licensed training data for its models. ⚖ Regulatory Evolution: Assurance systems for AI products will inevitably fail at times, underscoring the need for nuanced regulatory frameworks in privacy, cybersecurity, consumer protection, and competition law. This will include content moderation issues, as Apple recognises. 📚 AI Literacy: We need to foster proactive AI literacy, otherwise people will learn lessons about how these technologies work the hard way, like through scams, deception, or privacy breaches. Clearly we need to develop the research community, along with industry, but (as the AIRA report notes) we’re also going to need to build a talent pipeline for AI governance, audit and regulation too. I’m contacted frequently by young people looking for opportunities in this space, and I’m looking into what the Brainbox Institute can do to support this important area.

28

4 Comments

Thank you to ACS (Australian Computer Society) for supporting NAIDOC Week- a celebration of Aboriginal and Torres Strait Islander peoples. "It is also a time for reflection and for the tech sector to consider how it improves the representation of First Nations Australians." As part of NAIDOC Week, we shared Key insights from our webinar on "Generative AI and Indigenous Data: The Australian Perspective" hosted by #ACS, which brought to light several critical points regarding the intersection of emerging technologies and Indigenous data sovereignty. The webinar underscored the necessity of integrating Indigenous perspectives and knowledge systems into the development and deployment of AI technologies to ensure cultural sensitivity and ethical use.   "Importance of Indigenous Data Sovereignty" A recurring theme was the need to safeguard Indigenous data from exploitation. Professor Natalie Stoianoff, Director, IP Program, UTS Indigenous Knowledge Forum underscored the legal frameworks required to protect Indigenous cultural heritage, while Tui Raven, Author of the Guidelines for First Nations Collection Description, and co-investigator and cultural advisor for the First Inventors documentary highlighted the spiritual and cultural imperatives of data protection. There was a strong emphasis on the risks associated with data mining, which could potentially expose sensitive cultural information without proper consent and control.   "UNESCO's Role and Multilingualism" Jaco Du Toit, Chief of Section Universal Access to Information, from #UNESCO presented several global instruments and policies that can guide the engagement with Indigenous communities. Notably, he mentioned the 2003 Recommendation on the Promotion and Use of Multilingualism and Universal Access to Cyberspace. This recommendation is pivotal in fostering the creation and preservation of Indigenous knowledge, particularly in native languages, thereby ensuring that these communities are not mere consumers but active participants in the digital space.   Conclusion  The speakers urged for continued dialogue and collaboration among Indigenous communities, policymakers, and technologists to develop ethical AI frameworks that respect and protect Indigenous data. As a follow-up, participants are invited to engage with UNESCO's initiatives (https://lnkd.in/gnN3aFtF), including the global task force on digital empowerment and Indigenous languages. Your involvement can help shape a future where technology supports and enhances Indigenous knowledge and cultural heritage. https://lnkd.in/gsYpJSbj #AI ACM, Association for Computing Machinery #IITPSA -Institute of IT Professionals South Africa CIPS (Canada’s Association of I.T. Professionals) Computer Society of Zimbabwe Sociedade Brasileira de Computação - SBC Den norske dataforening (DND) Ordem dos Engenheiros Dataföreningen i Sverige #ifip Joyce Long Elana Huthnance Helen McHugh Cindy Chung

33

2 Comments

It's a privilege to be involved in cutting-edge applied research aimed at addressing real-world issues at Deakin University's #AppliedAI institute. https://lnkd.in/gC_YwadZ Focusing on #heatlh, #defence, #finance and more... * Improving Defense AI readiness * Utilizing machine learning to design functional molecules. * Helping surveillance operators identify the right footage to focus on. * #Computervision to aid in the early diagnosis of cerebral palsy in infants and continued commitment to resolve some of the hard core challenges faced by #industries & #societies today - #Optimization, #Robustness of AI, empathy in robotics and more... Kon Mouzakis Rajesh Vasa Olaf Pietschner Zhiwei Jiang Anirban Bose Shin(ichi) Tonomura Manoj KHERA Karl Mehta Dr Sunil Kumar Gupta

61

We are working on an intricate challenge in modern healthcare in #Australia to tackle the portability issue of causal CDSS from one healthcare provider to another healthcare provider by leveraging the meta-heuristic approach and AI. To mitigate this challenge, we have proposed an Adaptive Semantic Framework (ASF) which will do the needful adjustments in the existing CDSS according to a new environment. The abstract of our paper is given below: Clinical Decision Support Systems (CDSS) are a critical part of modern healthcare, assisting healthcare practitioners in making precise decisions. However, most of the existing CDSSs are static, which makes adaptation to a new environment or changes in the same environment challenging. This is where the Adaptive Semantic Framework (ASF) comes in. In a recent paper, a team proposed an ASF for CDSS adaptation to a changing environment. The framework includes various methods such as Knowledge-Based Systems (KBS) and Data-Driven Elements (DDE) to adjust the CDSS to new conditions. This framework addresses two significant challenges of CDSS: its maintenance and the issue of portability across different hospitals with varying data ecosystems. With the ASF, CDSS can now adapt to a new environment or local changes in the same environment, including different types of patients, different ecosystems, and varying guidelines. #CDSS #HealthcareTechnology #AdaptiveSemanticFramework #DataDrivenElements #KnowledgeBasedSystems #Research

7

✨V4E1 Spotlight✨ Fairness in AI models is a matter of growing importance in the field. Kejun Dai presents different frameworks for evaluating bias in AI models. This research involved learning each framework and running benchmark experiments. Read the full article here: https://lnkd.in/gKeSPbtD

3

MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) PhD candidate Andi Peng works alongside Professors Jacob Andreas and Julie Shah in the lab's Embodied Intelligence Group. Peng’s research generally focuses on deriving algorithms for learning from human feedback, whether that be in the form of querying for human data directly by asking them or by leveraging existing priors such as language models for representing human preferences. She said she is excited by the opportunities opened up by recent trends in pre-trained models for better capturing large, diverse preferences from humans and being able to better build robots and algorithms that learn from them. “Ultimately, we as researchers are building technology for human users — therefore, it’s incredibly important to capture the desired outcomes and behaviors that end users want,” Peng said. “I believe that regardless of how smart your AI gets in the future, there will always be an adjustment period where it needs to learn to adapt to its specific user’s preferences — therefore, it’s important to always center the end user in the formulation of the learning framework. Whether in language, vision, or robotics, this type of work will be important for ensuring that the agents we build are actually doing what people in the world want them to be doing.” Read more about Andi and her research: https://bit.ly/3PXsomf

25

1 Comment

Benchmarking Mathematics: The Best AI Models https://lnkd.in/gvDUVmJ5 For those following AI Benchmarking and interested fellow travellers, there is a new AI Mathematics benchmark from Epoch AI that dwarfs all other benchmarks (MMLU etc.). The video brief linked here is worth watching as the methodology is different and utilizes a gathering of the best mathematicians globally with longer term complex problems. The model benchmarks are also designed to help mathematicians work with longer term challenging problems (very long term difficult problems) as a co-intelligence. https://lnkd.in/gvDUVmJ5 #AIBenchmarking, #AIMathematics, #FutureofAI

1

A plug for the role of interdisciplinary education in the future of AI in a cameo appearance by yours truly in this TODAY article by Amanda Yeap on where Singapore stands in the global AI landscape.

21

2 Comments

Executive Summary The report “Australian Strategic Policy Institute’s Two-Decade Critical Technology Tracker: The Rewards of Long-Term Research Investment” provides an in-depth analysis of the benefits and insights gained from two decades of research on critical technologies. The report highlights the value of sustained #investment in #technology #research and its impact on strategic decision-making and #innovation. Key findings and recommendations are summarized below: 1. Long-Term Research Benefits: • Informed Decision-Making: Long-term research provides a comprehensive understanding of technology trends and their implications. This knowledge supports better strategic decisions and helps organizations anticipate and adapt to technological changes. • Strategic Advantage: Continuous investment in technology research enhances an organization’s ability to stay ahead of competitors by identifying emerging trends and assessing their potential impact on various industries. 2. Key Insights from Two Decades of Research: • Technology Evolution: The report tracks the evolution of critical technologies over the past 20 years, highlighting key advancements and shifts in technology landscapes. This historical perspective offers valuable insights into how technologies have developed and their future trajectories. • Impact on Security and Strategy: Insights gained from long-term research have been instrumental in shaping security strategies and technology policies. Understanding the dynamics of critical technologies helps in mitigating risks and leveraging opportunities for innovation. 3. Challenges and Considerations: • Sustaining Research Efforts: Maintaining long-term research initiatives requires ongoing funding and commitment. Ensuring sustained investment is crucial for continued success and relevance in a rapidly changing technological environment. • Adapting to Change: Organizations must be prepared to adapt their strategies based on evolving research findings. Flexibility and responsiveness to new insights are essential for maximizing the benefits of long-term research. 4. Recommendations for Future Research: • Continued Investment: Organizations should continue investing in long-term technology research to gain a strategic advantage and stay informed about emerging trends. • Collaborative Efforts: Collaboration with academic institutions, industry partners, and research organizations can enhance the quality and scope of research efforts. • Focus on Impact: Research should focus on technologies with significant potential impact, both in terms of innovation and security, to ensure that investments yield meaningful benefits. The report underscores the importance of long-term research in #technology, demonstrating that sustained #investment leads to valuable insights and #strategicadvantages. By maintaining a commitment to research, organizations can effectively navigate #technologicaladvancements and enhance their strategic positioning.

5

Great tutorial in #ALTA2024 Efficient training of small-scale #LLMs in resource constrained environments 👩🏻‍💻

4

Australia's Senate Select Committee on AI Adoption last night published Meta's responses to Senator Tony Sheldon's questions about the scope, sources, and legality of data used to train the company’s AI models, including whether public and business content, youth data, or pirated datasets (like Books3) were involved, what privacy safeguards and opt-out mechanisms are in place, and whether Meta complied with copyright laws, particularly in Australia. Meta's answers are, obviously, incredibly cautious and precisely vague (that's a PR strategy, I'm not being facetious). For AI nerds like me who've been following this, there's really not much that's new here. But here's some thoughts from me. Please comment to help confirm, or correct me if I'm wrong here: 🔵Retroactive data removal process — Meta says users can retroactively opt-out past publicly posted content from being used in *future* AI model training through a bulk process. I don't think this feature has been publicised or is widely known? 🔵Clarification on business content — Public posts from business accounts (including artistic, illustrative, or commercial content) are used in AI model training. I think this is the first confirmation that such data is part of the training dataset? But, also, hardly a surprise. 🔵Books3 — Meta continues to not deny its use of the Books3 dataset, a collection of pirated books, to train its LLaMA models. But we knew this already. 🔵Role of Robots.txt in AI scraping — Meta says it uses robots.txt as the standard protocol to respect websites' permissions for AI training. But, it said that this mechanism governs only *future* data collection, and retroactive removal of scraped content is not supported. I think that's a nuance not previously revealed? 🔵1m domains removed — Meta revealed that it filters out more than a million domains from its training datasets due to privacy risks. I'd sure love to know what sites are on *that* list?🤔 🔵De-identification of PII — There's a bit of detail on Meta's methods to remove identifiers from both first-party and third-party datasets, including structured identifiers like Meta user IDs and usernames; but nothing that's really groundbreaking here. 🔵Australian Copyright law — Obviously, Meta hasn't been drawn on its views about Australian copyright law. It's not an Australian company and attempts to drag it into Australian jurisdiction seem to me like a distraction. (Of course, the EU has a very different view on this, but they also have the legislation to back it up.)

25

14 Comments