This product was not featured by Product Hunt yet. It will not be visible on their landing page and won't be ranked (cannot win product of the day regardless of upvotes).
Coincidex
Transfer-aware, replay-free continual learning for PyTorch
Most continual-learning methods either forget (fine-tuning) or isolate and forfeit transfer (fixed masks, adapters). Coincidex does both: similar tasks share sub-networks for forward transfer, dissimilar ones stay isolated so old skills survive — with no replay buffer. A tiny, drop-in PyTorch layer (torch + numpy, CPU-friendly) that learns task similarity from context. It's also honest: reproducible benchmarks and a clear map of where it works and where it doesn't. Apache-2.0 license.
Hi everyone,
I'm Rakib. Coincidex is a small PyTorch layer for continual learning. The short version: it lets a model keep learning new tasks while transferring what it can to related ones and holding on to the old ones, and it does that without a replay buffer.
The problem it addresses: when you train a model on a sequence of tasks or domains, plain fine-tuning forgets the earlier ones. The usual fix is to isolate each task (fixed random masks, or one adapter per task). That stops the forgetting, but it throws away transfer, so every new task starts from scratch even when it's close to something the model already learned.
Coincidex is the middle ground. It uses a context signal to route similar tasks to overlapping parts of the network so they share and transfer, and dissimilar tasks to separate parts so they don't overwrite each other. The gate is hard on/off, so a switched-off unit gets no gradient at all and its old weights are simply frozen. Nothing stored.
Two things I want to be upfront about:
It's honest about its limits. I tested it by trying to break it, and the repo has a plain section on where it works and where it doesn't. It works for domain and covariate shift, and it helps most on smaller or weaker models. It does not help for class-incremental problems (genuinely new classes appearing over time), and it adds little when a large frozen model already handles your inputs well. Better you hear that from me than find out later.
It's small and easy to try. torch and numpy only, runs on a laptop, drop-in nn.Module, with baselines and reproducible benchmarks included. It's early (v0.2, research/beta) and Apache-2.0.
Repo: https://github.com/rakib-nyc/coi...
If you're dealing with catastrophic forgetting, concept drift, or a stack of per-customer or per-domain models you keep retraining from scratch, I'd like to hear about it. Feedback and PRs welcome.
Thanks for taking a look.
Rakib
one thing that would make this much more useful for me — a small visualisation of the learned task-similarity matrix as it evolves across tasks. even just a heatmap saved per checkpoint would help me trust the routing decisions and debug when transfer isnt kicking in. could be optional and opt-in so it doesnt slow anything down.
the context-driven similarity learning without a replay buffer is genuinely clever, that alone usually forces a tradeoff you avoided. appreciate the honest breakdown of where it falls short too, that kind of transparency is rare.
dropped it into a small continual vision setup and it just worked without me hand-tuning masks. the honesty about failure modes is refreshing, more libraries should do that.
finally someone tackling continual learning without the replay buffer baggage, kind of refreshing actually. the task similarity layer is tiny but the no replay + honest benchmarks angle is what got me, though i'd want to see it on tougher task sequences before fully buying in
Dropped it into a small continual-learning setup and was surprised how little code it took — basically one layer swap. Love that it ships without a replay buffer and admits where it falls short instead of overpromising.
About Coincidex on Product Hunt
“Transfer-aware, replay-free continual learning for PyTorch”
Coincidex was submitted on Product Hunt and earned 9 upvotes and 6 comments, placing #151 on the daily leaderboard. Most continual-learning methods either forget (fine-tuning) or isolate and forfeit transfer (fixed masks, adapters). Coincidex does both: similar tasks share sub-networks for forward transfer, dissimilar ones stay isolated so old skills survive — with no replay buffer. A tiny, drop-in PyTorch layer (torch + numpy, CPU-friendly) that learns task similarity from context. It's also honest: reproducible benchmarks and a clear map of where it works and where it doesn't. Apache-2.0 license.
Coincidex was featured in Open Source (68.6k followers), Developer Tools (516.2k followers), Artificial Intelligence (474.2k followers) and GitHub (41.3k followers) on Product Hunt. Together, these topics include over 223.9k products, making this a competitive space to launch in.
Who hunted Coincidex?
Coincidex was hunted by Rakib. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
Want to see how Coincidex stacked up against nearby launches in real time? Check out the live launch dashboard for upvote speed charts, proximity comparisons, and more analytics.
Rakib