Recently, @OpenAI, @GoogleDeepMind, @ElevenLabs, and @AnthropicAI (support.claude.com/en/articl…) are all pushing #watermarking and #provenance forward. These efforts raise an important question: If model owners can watermark the content generated by their AI models for provenance and traceability, shouldn’t data owners be able to do likewise and know whether their personal content is used by the AI models? In other words, alongside asking, “Is this content generated by that AI model?” shouldn’t we also be able to ask, "Is that AI model trained on my personal content?” Back in 2023 (i.e., well before data provenance for LLM training became a mainstream conversation), we were already thinking seriously about this equally important problem. If the personal content of a data owner could be watermarked, its downstream use could be traced, including whether it was used to train an LLM. At that time, existing text watermarking methods were not robust nor practical enough for protecting arbitrary data at scale. That motivated us to develop a text watermarking framework, Waterfall, specially designed for data owners. Just like how watermarking can give model owners visibility into whether their generated content is being used, Waterfall can give data owners visibility into whether their personal content is being used. Check out our post from back then: xitter.cf/bryanklow/status/18512…

🚨 ATTENTION ALL DATA OWNERS: YOUR DATA IS NOT SAFE 📣 Your data might be used as LLM training data, and you might not even know it! So, once an LLM is deployed, how do you prove it? Here are three questions every data owner/auditor should be able to answer: • “Was my data used to train this LLM?” -> membership audit • ⁠”Can the LLM outputs be attributed back to their source (i.e., the data owner )?” -> source attribution • ⁠”Has the influence of my data been removed?” -> machine unlearning The problem: Today, these questions are studied in isolation with different methods and datasets 💔. A method that performs well on one can quietly fail the others. So, for researchers working on data provenance, we have something for you 🫵 Introducing Watershed, the first unified benchmark for an end-to-end data provenance pipeline from watermarking a dataset all the way to black-box auditing of an LLM. We introduce 1️⃣ shared protocol across 5 watermarking schemes, MIA baselines, and release watermarked datasets on 2 domains. Our key findings: A method's ranking at one stage doesn't predict its ranking at the next; some types of watermarks persist through training better than others; MIAs work great but are not robust to paraphrase attacks. 🤔 Catch our poster @icmlconf #ICML2026’s AI4GOOD workshop (trustworthy-ai-for-good.gith…) on 10 July at Grand Ballroom 103! This work was co-led by @john_r_himawan @greglau and our amazing collaborators @lululu0082 @xinyuan3142 @bryanklow 🫡. 📜 Paper: openreview.net/forum?id=Fb7t… 🤗 Huggingface: huggingface.co/collections/G… ⭐️ GitHub: github.com/johnrhimawan/wate…

🎼 🎵 🎶... Time to party @icmlconf #ICML2026? Let's go! Wait, I'm the last to depart from 🇸🇬 ??? Don't miss out on the 𝐍𝐔𝐒 𝐱 𝐉𝐚𝐧𝐞 𝐒𝐭𝐫𝐞𝐞𝐭 𝐒𝐮𝐦𝐦𝐢𝐭 𝐨𝐧 𝐒𝐞𝐥𝐟-𝐄𝐯𝐨𝐥𝐮𝐭𝐢𝐨𝐧 𝐨𝐟 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭𝐬 𝐚𝐧𝐝 𝐌𝐮𝐥𝐭𝐢𝐚𝐠𝐞𝐧𝐭 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 (9 Jul)! See link below: xitter.cf/bryanklow/status/20696… #LLMs #AgenticAI #AIAgents #AgenticMemory #DataSelection #ShapleyValue #MachineUnlearning #BayesianOptimization #SpeculativeDecoding

🚀 What should we evolve next in AI agents? 🤖✨ As #AgenticAI rapidly moves from a single-agent system 🤖 to a team of agents 🤖🤖🤖 that remember, coordinate, search, optimize, and interact in increasingly complex environments, this question is particularly urgent. 🔥 To bring researchers together around this emerging direction and following the success of the NUS x DSO x AWS Symposium on Agentic AI Meets Autonomous Agents and Multiagent Systems held during @RealAAAI #AAAI2026 (xitter.cf/bryanklow/status/20145…), the National University of Singapore (NUS @NUSingapore) and Jane Street (@JaneStreetGroup) are pleased to host and welcome you to: NUS x Jane Street Summit on Self-Evolution of AI Agents and Multiagent Systems 📅 Date: 9 July 2026 ⏰ Time: 14:30 - 19:00 📍 Location: Seoul 🇰🇷 (Held in conjunction with @icmlconf #ICML2026 and close to the conference venue) ⚠️ Capacity is limited, and registrations will be reviewed and confirmed by the organizers. Secure your spot early! 👇 🔗 Register here: luma.com/rzkweojj We invite researchers at the forefront of Agentic AI to join us in - shaping key research questions, - exploring new applications, - identifying potential collaborators, and - influencing the technical discourse around what to evolve in agents. 🎯 The keynote speeches and roundtable discussions will be centered on how future AI agents can evolve across 3 key dimensions: 🧠 Memory (e.g., MEM1 xitter.cf/bryanklow/status/20157…, MeMo xitter.cf/bryanklow/status/20576…) 👥 Collaboration & Coordination (e.g., CORAL xitter.cf/bryanklow/status/20346…) 🔍 Optimization & Search (e.g., CORAL xitter.cf/bryanklow/status/20346…) 📝 Goal: The outcome of the Summit is a position or survey paper to be submitted to a top AI venue. Participants at the Summit will be invited to co-author this paper! ✍️📚 ☕ Food and drinks will be available during the Summit. #AIAgents #MultiAgentSystems #LLM #LLMs

[1/5] 🚨 How do you prove that an #LLM company actually deleted your private or copyrighted data upon your request? 🕵️‍♂️ Spoiler: right now, you really can't! 😱 Existing LLM unlearning metrics are not applicable in the real world. 😬 Thrilled to present WaterDrum (#ICLR2026)! We use text #watermarks to finally measure what an LLM has truly "forgotten."💧🥁 This work was co-led by @lululu0082 @xinyuan3142 @greglau in collaboration with @nhungbui1299 @RachaelSim2 John Russell Himawan, Fanyu Wen, Chuan-Sheng Foo, See-Kiong Ng, @bryanklow. 📄 Paper: openreview.net/pdf?id=5GVfne… Catch our poster @iclr_conf on 23 Apr 10:30AM Pavilion 4 P4-#4210! 🇧🇷 See the thread below 🧵👇

[1/4] You want to fine-tune an #LLM. 1️⃣ What's the best data mixture? 2️⃣ What LoRA config. to use? The answers to the above depend on each other. 🐔🥚 Our team (with @ZhiliangChen94 @alfredleongwl @ShaoYongOng @apivich_h @greglau et al.) calls this the Chicken-and-Egg Dilemma of LLM fine-tuning — and we just solved it. 📄Paper: arxiv.org/pdf/2602.08351 📍Poster: @iclr_conf #ICLR2026 DATA-FM Workshop, 26 Apr, Riocentro Room 203 A+B

[1/3] 🤔An interesting and practical question: How can we find the optimal #LLM training data mixture that maximizes a free-form downstream task metric? For instance, what fine-tuning data mixture should we use to maximize same-demographic user ratings ⭐ across our chatbots? Our #ICLR2026 work (with @ZhiliangChen94 @greglau Chuan-Sheng Foo) called DUET interleaves #BayesianOptimization and #DataSelection to automatically discover the best data mixture that maximizes any free-form downstream feedback, without manually searching through countless combinations. 📄Paper: arxiv.org/abs/2502.00270 📅Catch us at @iclr_conf 🇧🇷Poster Session 3 Fri Apr 24 10:30AM Pavilion 3 P3-#305. More below👇.

When a company claims that your personal data has been removed from their model, have you ever wondered whether they've indeed done so? 🤔 If a new paper on arXiv claims that its proposed #MachineUnlearning algorithm can unlearn your personal data from an #LLM, how do we know if it can indeed do so? 🛢️WaterDrum is the first data-centric LLM unlearning metric based on watermarking that is calibrated, requires no retraining, works for blackbox models and when forget/retain sets have similar data. Let the (Water)Drums roll at Rio! @iclr_conf #ICLR2026

Given a single model, how do we improve an #LLM’s reasoning performance with limited resources 💻 and inference time ⌛️? Can a smaller 1.5B model outperform a 7B model without incurring long inference time from sequential queries? In the work of @_Hu_Wenyang @greglau et al., we introduce the framework called Dipper to create #LLMs ensembles from an optimized set of diverse reasoning prompts to improve performance. Dipper runs queries in parallel with a prompt optimization method inspired by Determinantal Point Processes (DPP), making it super fast ⏩ and effective. Furthermore, Dipper can work with LLM APIs without model access 📦! With Dipper, we demonstrated how a small ensemble of just three 1.5B models can outperform a 7B model on a range of math and non-math reasoning tasks, while taking almost the same inference time and just < 3x compute for a normal query thanks to accelerated batch inference methods 😱 ! Find out more at @emnlpmeeting #EMNLP2025 (Poster 4492) at Hall C, Session 11, Nov 6 at 16:30! Paper: arxiv.org/abs/2412.15238

Inspired by the success of inverse problems in uncovering fundamental scientific laws, our position paper, which is accepted to @emnlpmeeting #EMNLP2025 (findings), argues that inverse problems can be used to efficiently uncover underlying scaling laws for #LLMs that help the building of LLMs to achieve a desirable performance with significantly better cost-effectiveness. Joint work with @arun_v3rma @WuZhaoxuan @BobbyZhouZijian @xiaoqiang_98 @ZhiliangChen94 @RachaelSim2 @ray_qiaorui @JingtanW @nhungbui1299 @greglau @_Hu_Wenyang @xinyuan3142 Zi-Yu @ZhaoZitong00 @michael_xinyi @apivich_h See-Kiong.

Joint work with @_Hu_Wenyang @greglau et al. on Dipper (Diversity in Prompts for Producing #LLM Ensembles in Reasoning tasks) is accepted to @emnlpmeeting #EMNLP2025 (main)!

When a company claims that your personal data has been removed from their model, have you ever wondered whether they've indeed done so? 🤔 If a new paper on arXiv claims that its proposed #MachineUnlearning algorithm can unlearn your personal data from an #LLM, how do we know if it can indeed do so? 🛢️WaterDrum is the first data-centric LLM unlearning metric based on watermarking that is calibrated, requires no retraining, works for blackbox models and when forget/retain sets have similar data. Find out more about Waterdrum from @greglau at our poster today (18 July) in the #ICML2025 Machine Unlearning for #GenerativeAI (MUGen @icmlconf) workshop during 3-3.45pm in West Meeting Room 202-204! Joint work with @lululu0082 xinyuan @greglau nhung @RachaelSim2 fanyu chuansheng seekiong.

🤔How can we reliably use #Multimodal Large Language Models (#MLLMs) in practical settings? With multiple modalities come more challenges in managing uncertainty and mitigating errors where responses may seem plausible but are incorrect 😱. Introducing ⚖️UMPIRE, an #UncertaintyQuantification framework for MLLMs that estimates task instance uncertainties at inference-time without external tools or additional training. Inspired by #DeterminantalPointProcess (#DPP), the UMPIRE metric is based on a global measure of the semantic volume formed by sampled responses, adjusted by local measures of each sample’s coherence with the multimodal query. Find out more about UMPIRE from @greglau at poster tomorrow (19 July) in the @icmlconf #ICML2025 #R2-FM Reliable and Responsible Foundation Models workshop during 12-1pm and 4-5pm in West Ballroom C! Joint work with @greglau and hieu. Paper: openreview.net/pdf?id=2UYZHv…

Obtaining 🔎 interpretable symbolic equations is critical for scientific understanding and decision making in high-stakes environment. In fact, many such real-world applications (e.g., interactive/adaptive experiments, real-time decision support) demand rapid results within ⏳seconds. Hence, we ask❓: how can we rapidly discover accurate and parsimonious symbolic equations from data? In 📖README: Rapid #EquationDiscovery using #MultimodalEncoders, we develop an efficient #Multimodal foundation model for rapid equation discovery with 3 key insights: 1.⁠ ⁠Use image plots rather than raw numerical data to efficiently extract key trends; 2.⁠ ⁠Make use of pre-trained image and text encoders to efficiently train foundation model for equation discovery; 3.⁠ ⁠Develop an query-efficient optimization approach for efficient inference (Grey Wolf Algorithm x #BayesianOptimization). Find out more about README from @greglau at our poster today (18 July) in the @icmlconf #ICML2025 AI4MATH workshop during 10:50am-12.20pm in West Ballroom C! Joint work with @greglau yueran zi-yu @apivich_h @ruthchewing.

GLOW.AI (comp.nus.edu.sg/~lowkh/) will be @icmlconf #ICML2025: @ray_qiaorui @JingtanW @greglau @_Hu_Wenyang @ruthchewing! More details on each work later. Interested to apply for a faculty position or Ph.D. in AI at @NUSComputing , ping me on Whova to meet up. See you soon! #DataSelection #DataCentricAI #LLMs #LLM #FederatedLearning #BayesianOptimization #MachineUnlearning #RLHF #SymbolicRegression

Introducing WaterDrum🛢️, the first data-centric #LLM #unlearning metric that leverages robust text #watermarking💧to provide an effective, practical, and resilient way to evaluate LLM unlearning performance😉! (1/n) #MachineUnlearning #LLMs

@greglau and @apivich_h are getting PIED @iclr_conf #ICLR2025 🦶🥧🤣

🧪 A key challenge to solving #InverseProblems, especially in science and engineering settings, is limited data and compute for simulations. 🥧 The @iclr_conf work of @apivich_h @greglau , PIED, is an #ExperimentalDesign framework for #PDE inverse problems that efficiently optimizes for the best observation inputs. 📌 PIED utilizes #PINNs to perform forward simulation and to solve the inverse problems, which allows for higher efficiency over numerical simulators and allow for meshless and differentiable simulations. 🤖 Using PINNs along with differentiable observation selection criteria, PIED selects the optimal design parameters for one-shot deployment, while exploitation parallel computation and gradient-based optimization methods. 📈 These benefits allow PIED to outperform existing ED benchmarks in IPs for both finite-dimensional and function-values inverse parameters, as demonstrated in various settings including on real experimental data. Check out our #ICLR2025 Poster #28 on 26 Apr in Poster Session 6 Sat 3pm Hall 3 + Hall 2B. Paper: openreview.net/forum?id=w7P9… GitHub: github.com/apivich-h/pied