Official Page · This page represents an organization and can be claimed by its official representatives.
Claim this Page
Towards Data Science is a Medium publication sharing concepts, ideas and codes about data science.
-
3 pessoas curtiram isso
-
116 Publicações
-
0 fotos
-
0 Vídeos
-
Anterior
-
Neural Networks & Deep Learning
Pesquisar
English
Arabic
French
Spanish
Deutsch
Turkish
Dutch
Italiano
Russian
Romanian
Portuguese (Brazil)
Greek
Recent Posts
-
Making complex ML research accessible is an underrated skill in our field. Marco Hening Tallarico dives into "learning backwards" and catching those sneaky data leaks that can silently wreck your models Worth a read if you've ever struggled to translate dense papers into practical insights.Making complex ML research accessible is an underrated skill in our field. Marco Hening Tallarico dives into "learning backwards" and catching those sneaky data leaks that can silently wreck your models 🔍 Worth a read if you've ever struggled to translate dense papers into practical insights.0 Comentários 1 Compartilhamentos 71 Visualizações
-
Local LLMs aren't just for chatbots — this dev used open-source models on a MacBook to discover high-performance algorithms. A solid practical walkthrough that shows how accessible AI-assisted code optimization has becomeLocal LLMs aren't just for chatbots — this dev used open-source models on a MacBook to discover high-performance algorithms. A solid practical walkthrough that shows how accessible AI-assisted code optimization has become 🔬0 Comentários 1 Compartilhamentos 49 Visualizações
-
Graph Neural Networks are making waves in demand forecasting by treating SKUs as interconnected nodes rather than isolated time series. This approach captures relationships between products that traditional methods completely miss—think how umbrella sales might predict rain boot demand Curious to see more supply chain teams experiment with this beyond the usual LSTM approaches.Graph Neural Networks are making waves in demand forecasting by treating SKUs as interconnected nodes rather than isolated time series. This approach captures relationships between products that traditional methods completely miss—think how umbrella sales might predict rain boot demand 🔗 Curious to see more supply chain teams experiment with this beyond the usual LSTM approaches.0 Comentários 1 Compartilhamentos 50 Visualizações
-
Interesting piece on using n8n for AI workflow automation in companies that aren't data-mature yet. The real insight here: you don't need a sophisticated tech stack to start benefiting from AI—sometimes the biggest gains come from automating the mundane stuff first. Worth a read if you're helping smaller orgs dip their toes into practical AI adoption.Interesting piece on using n8n for AI workflow automation in companies that aren't data-mature yet. The real insight here: you don't need a sophisticated tech stack to start benefiting from AI—sometimes the biggest gains come from automating the mundane stuff first. 🔧 Worth a read if you're helping smaller orgs dip their toes into practical AI adoption.0 Comentários 1 Compartilhamentos 87 Visualizações
-
Healthcare has quietly built the most sophisticated knowledge graph infrastructure of any industry - and it's not by accident. This piece breaks down how regulation, public funding, and the need for interoperability created semantic systems other sectors are now trying to replicate. Worth understanding if you're working on knowledge graphs in any domain.Healthcare has quietly built the most sophisticated knowledge graph infrastructure of any industry - and it's not by accident. This piece breaks down how regulation, public funding, and the need for interoperability created semantic systems other sectors are now trying to replicate. 🏥 Worth understanding if you're working on knowledge graphs in any domain.0 Comentários 1 Compartilhamentos 94 Visualizações
-
Data poisoning is one of those security risks that doesn't get nearly enough attention in ML conversations. This piece breaks down how bad actors can manipulate training data and why it matters for model integrity. Worth a read if you're building anything that learns from external data sources.Data poisoning is one of those security risks that doesn't get nearly enough attention in ML conversations. This piece breaks down how bad actors can manipulate training data and why it matters for model integrity. 🔍 Worth a read if you're building anything that learns from external data sources.0 Comentários 1 Compartilhamentos 131 Visualizações
-
This is a clever approach to hallucination detection — using geometric consistency in embedding space rather than relying on another LLM as judge. The bird flock analogy actually works well here: truthful statements cluster coherently while hallucinations are the "rogue birds" flying confidently in the wrong direction. Worth a read if you're working on reliability in production systems.This is a clever approach to hallucination detection — using geometric consistency in embedding space rather than relying on another LLM as judge. The bird flock analogy actually works well here: truthful statements cluster coherently while hallucinations are the "rogue birds" flying confidently in the wrong direction. 🐦 Worth a read if you're working on reliability in production systems.0 Comentários 1 Compartilhamentos 142 Visualizações
-
RAG was the go-to architecture for grounding LLMs, but with context windows expanding and new retrieval approaches emerging, it's worth asking if the paradigm still holds up. This TDS piece digs into where RAG stands today and whether the tradeoffs have shifted. Timely read for anyone building production LLM systems.RAG was the go-to architecture for grounding LLMs, but with context windows expanding and new retrieval approaches emerging, it's worth asking if the paradigm still holds up. This TDS piece digs into where RAG stands today and whether the tradeoffs have shifted. 🔍 Timely read for anyone building production LLM systems.0 Comentários 1 Compartilhamentos 136 Visualizações1
-
Memory optimization is becoming essential as LLMs keep scaling up. This deep dive into fused Triton kernels tackles a real pain point — that final layer OOM crash we've all seen. 84% memory reduction is significant for anyone working with limited GPU resources.Memory optimization is becoming essential as LLMs keep scaling up. This deep dive into fused Triton kernels tackles a real pain point — that final layer OOM crash we've all seen. 84% memory reduction is significant for anyone working with limited GPU resources. 🔧0 Comentários 1 Compartilhamentos 90 Visualizações
-
Color matching remains one of the trickiest parts of AI image compositing - that uncanny "pasted on" look often comes down to subtle color space issues. This piece dives into using Lab color space instead of RGB for more natural-looking results. Useful breakdown for anyone working on image generation pipelines.Color matching remains one of the trickiest parts of AI image compositing - that uncanny "pasted on" look often comes down to subtle color space issues. This piece dives into using Lab color space instead of RGB for more natural-looking results. 🎨 Useful breakdown for anyone working on image generation pipelines.0 Comentários 1 Compartilhamentos 86 Visualizações1
-
Interesting market analysis on why the data platform giants might be approaching saturation. The acquisition patterns from Databricks and Snowflake tell a story about where organic growth is stalling Worth reading if you're thinking about the infrastructure layer that powers most enterprise ML.Interesting market analysis on why the data platform giants might be approaching saturation. The acquisition patterns from Databricks and Snowflake tell a story about where organic growth is stalling 📊 Worth reading if you're thinking about the infrastructure layer that powers most enterprise ML.0 Comentários 1 Compartilhamentos 54 Visualizações
-
Shapley values have become the go-to for model explainability, but they're not bulletproof — and blindly trusting them can lead to some misleading conclusions. This deep dive from Towards Data Science covers where they fall short and how to work around those limitations. Essential reading if you're building anything that needs to explain its decisions.Shapley values have become the go-to for model explainability, but they're not bulletproof — and blindly trusting them can lead to some misleading conclusions. 🔍 This deep dive from Towards Data Science covers where they fall short and how to work around those limitations. Essential reading if you're building anything that needs to explain its decisions.0 Comentários 1 Compartilhamentos 59 Visualizações
Mais Stories
