# multi-task (blogs) — RSS Amplifier Recent posts from the 3 feeds in the RSS Amplifier directory that cover multi-task. Page: Feed: --- ## [Multi-Task Learning: Calm Under Visual Pressure](https://michaelmallari.bitbucket.io/case-study/multi-task-learning-calm-under-visual-pressure/) _2026-01-18 · AI-First Product Management_ How condition-aware perception improves reliability, safety, and decision-making in complex environments. ## [Multi-Task Vision: When Bad Weather Throws a Wrench](https://michaelmallari.bitbucket.io/research-paper/multi-task-vision-when-bad-weather-throws-a-wrench/) _2026-01-17 · AI-First Product Management_ How RobuMTL improves multi-task computer vision reliability under degraded and mixed real-world conditions. ## [Repurchase Timing: Counting Days Without Overthinking](https://michaelmallari.bitbucket.io/case-study/repurchase-timing-counting-days-without-overthinking/) _2026-01-17 · AI-First Product Management_ How disciplined prediction, smart evaluation, and customer control turn better timing into a competitive advantage. ## [Inter-Purchase Intervals: When More Context Misses the Point](https://michaelmallari.bitbucket.io/research-paper/inter-purchase-intervals-when-more-context-misses-the-point/) _2026-01-16 · AI-First Product Management_ Why inter-purchase interval prediction favors precision models over language models, and what “good enough” timing really means. ## [Signal Strength: Getting Agents on the Same Wavelength](https://michaelmallari.bitbucket.io/case-study/signal-strength-getting-agents-on-the-same-wavelength/) _2025-12-05 · AI-First Product Management_ Why structured communication and well-designed feedback loops turn fragmented AI workflows into trustworthy multi-agent solutions. ## [Partial Info, Full Drama: When Agents Just Don’t “Get the Message”](https://michaelmallari.bitbucket.io/research-paper/partial-info-full-drama-when-agents-just-dont-get-the-message/) _2025-12-04 · AI-First Product Management_ AsymPuzl shows why effective signaling and shared understanding are critical to trustworthy, scalable multi-agent AI systems. ## [Improving DeepSeek R1 in Math](https://hav4ik.github.io/improving-deepseek-r1/) _2025-04-18 · Chan Kha Vu 🇺🇦_ I joined a team and we trained 7B and 14B math reasoning models based on DeepSeek-R1-Distill using SFT and GRPO. Our 14B model achieved 75.8% Maj@32 on AIME’25 ( +8.7% improvement), and our 7B model reached 65.8% Maj@32 ( +7.5% ). Here is what I’ve learned. ## [Are Reasoning Abilities Low-Rank?](https://hav4ik.github.io/is-reasoning-low-rank/) _2025-02-17 · Chan Kha Vu 🇺🇦_ Turns out, the RL training process of DeepScaleR-1.5B introduced only low-rank changes to its base model, DeepSeek-R1-Distill-1.5B. ## [Learning to Rank in Web Search](https://hav4ik.github.io/learning-to-rank/) _2024-09-08 · Chan Kha Vu 🇺🇦_ Learning to Rank (LTR) is a core component of any recommendation system. It is the algorithm that forms the final list of items to be shown to the user. This blog post is a comprehensive introduction to the basics of LTR and Unbiased LTR. Hopefully, it will give you enough context to build your own models or to understand more recent research in the field.