Not sure of the exact title? Can't recall the author? Just describe what you remember — ilastik, that ML bio paper, the one with the weird figure. We find it.
Fuzzy match
Finds papers from partial names, vague descriptions, or misremembered keywords
Surface your notes in context
Surfaces your annotations, critiques, and highlights alongside the paper
Last left off
Picks up exactly where you stopped — your last reading position and open questions
Instant results
No boolean operators, no exact quotes — natural language across your entire library
You read dozens of papers a month. Remembering exact titles or authors is unrealistic — your search shouldn't require it.
"stik ml bio" matches ilastik — a machine-learning bioimage tool. We understand intent, not just spelling.
Your critique, your highlights, your questions — all pulled up alongside the paper the moment you find it.
Your last note, open question, or reading position is shown right in the result — no digging required.
Search across your entire uploaded library at once — papers, notes, annotations, and chat history.
The AI reformulates your vague prompt into a precise search — then explains what it found and why.
Results show related papers and concepts from your knowledge graph — not just the exact match.
Don't remember the exact title? Just describe what you're looking for — the way you'd ask a colleague. We match your fuzzy memory to the right paper and surface your notes on it instantly.
Paper
Berg, S., Kutra, D., Kroeger, T. et al.
We present ilastik, an easy-to-use interactive tool that brings machine-learning-based (bio)image analysis to end users without substantial computational expertise. It contains pre-defined workflows for image segmentation, object classification, counting and tracking. Users adapt the workflows to the problem at hand by interactively providing sparse training annotations for a nonlinear classifier. ilastik can process data in up to five dimensions (3D, time and number of channels). Its computational back end runs operations on-demand wherever possible, allowing for interactive prediction on data larger than RAM. Once the classifiers are trained, ilastik workflows can be applied to new data from the command line without further user interaction. We describe all ilastik workflows in detail, including three case studies and a discussion on the expected performance.
Your note
Strengths
Accessible interactive ML for bioimage analysis without coding. Sparse annotation (scribbles) lowers labeling burden. Supports segmentation, object classification, counting, and tracking across up to 5D data (3D + time + channels). Out-of-core, on-demand computation enables interactive prediction on datasets larger than RAM. Trained workflows can be exported and run headlessly via command line for reproducible batch processing.
Research Directions
• Active learning for interactive annotation: uncertainty-guided scribble suggestions to reduce labeling time.
• Self-supervised pretraining for 3D microscopy to reduce annotation burden.
• Hybrid deep feature integration: plug pretrained CNN/ViT features into ilastik classifier pipeline.
• Multi-resolution, coarse-to-fine inference for terabyte-scale datasets.
• Uncertainty-aware overlays to guide user corrections in real time.
• Systematic evaluation of annotation granularity (clicks vs scribbles vs outlines).
• Cross-dataset transfer and domain adaptation benchmarking.
• Standardized 5D interactive segmentation benchmark suite.
Concerns / Gaps
Limited exploration of model uncertainty calibration. Feature engineering remains largely hand-crafted. Generalization across modalities unclear. Human-in-the-loop efficiency not rigorously quantified. Scalability tradeoffs vs deep end-to-end models underexplored.
Last left off
Want to examine classifier type (random forest?) and feature stack in detail; compare to modern deep segmentation baselines and evaluate computational complexity scaling with dataset size.
Upload your library once. Search it forever — the way you actually think.
Join waitlist