Metadata Archaeology: Unearthing Data Subsets by Leveraging Training Dynamics | Cohere | Research | TechiSeek

AddAdd your Listings

Research

Metadata Archaeology: Unearthing Data Subsets by Leveraging Training Dynamics

Modern machine learning research relies on relatively few carefully curated datasets.

Research Cohere

ResearchThe Culture Funnel: You can’t align what isn’t in the dataCohere ResearchImproving Policy Learning via Language Dynamics DistillationCohere ResearchDiversify and Conquer: Diversity-Centric Data Selection with Iterative RefinementCohere ResearchLLM See, LLM Do: Guiding Data Generation to Target Non-Differentiable ObjectivesCohere ResearchPrioritized Training on Points that are Learnable, Worth Learning, and Not Yet LearntCohere ResearchNexus: Specialization meets Adaptability for Efficiently Training Mixture of ExpertsCohere

EventsCloud Security Trends & Challenges: Complete GuideCybersecurity Exchange LearnHow to use 1Password's Travel Mode on your phone, tablet, and laptop1password BlogsSimple Image Placeholders with SVG | CSS-TricksCss Tricks EventsMotherDuck + DuckDB MeetupMotherduck BlogsYour Obsidian Vault Can Now Run SQL (and Your Agent Can Read It)Motherduck Eventswith MotherDuckMotherduck

LearnThe Database Inside Your Lakehouse: A DuckLake Architecture Deep DiveMotherduck BlogsSelf-hosting DuckDB: the road to productionMotherduck EventsA new paradigm for data visualization with just SQL + MarkdownMotherduck BlogsUnion and MotherDuck's Orchestrated Approach to Advanced AnalyticsMotherduck EventsWe Classified 100,000 Rows in 40 Seconds: Introducing prompt_jev()Motherduck EventsSmall Data SF: Keynotes and SessionsMotherduck EventsAI Native Summit 2024Motherduck EventsStreaming Kafka Data into MotherDuck with Estuary FlowMotherduck