Doctoral defence of Federico Malato, MSc, 15.12.2025: Enhancing decision making with retrieval-learning hybrid agents MSc Federico Malato's doctoral dissertation explores a novel approach to augment decisions made by autonomous agents via active recall of past experiences.

Belated congratulations to Dr Federico Malato, one of the most active members of our StatML group, on earning his PhD on 15 Dec 2025! 👏🚀🎉

His dissertation explores retrieval learning hybrid agents: using a memory module + search to actively recall past experiences and improve decision making.

Generalizable speech deepfake detection via meta-learned LoRA Reliable detection of speech deepfakes (spoofs) must remain effective when the distribution of spoofing attacks shifts. We frame the task as domain generalization and show that inserting Low-Rank Adap...

🧵3/3:

2) “Generalizable speech deepfake detection via meta-learned LoRA” by Laakkonen, Kukanov, Hautamäki arxiv.org/abs/2502.108...

Speech deepfake detection under attack shift: LoRA adapters + meta-learning (MLDG) to learn transferable cues rather than overfitting to specific spoofing methods.

Targeted Fine-Tuning of DNN-Based Receivers via Influence Functions We present the first use of influence functions for deep learning-based wireless receivers. Applied to DeepRx, a fully convolutional receiver, influence analysis reveals which training samples drive b...

🧵2/3:

1) “Targeted Fine-Tuning of DNN-Based Receivers via Influence Functions” by Tuononen, Penttinen, Hautamäki (StatML) arxiv.org/abs/2509.15950

Influence functions pinpoint the training samples behind bit decisions, enabling targeted fine-tuning that improves BER (single-target > random).

🧵 ICASSP 2026 update: two papers involving StatML members have been accepted. 🎉

One on targeted fine tuning for DNN based wireless receivers using influence functions, and one on generalizable speech deepfake detection via meta learned LoRA.

Huge congratulations to all authors! 🙌

From UEF StatML to #NeurIPS 2025 in San Diego 🚀 Federico Malato is presenting together with Ville Hautamäki their poster “Zero shot World Models via Search in Memory”. Congratulations to the authors and thanks to everyone who stops by the poster 😊

At AI-DOC today: Laakkonen presenting the StatML project conducted by Laakkonen, Kukanov and Hautamäki 😊 A solid contribution from our StatML team 👏🏽🚀

#Deepfake
#AudioDeepfakes
#DeepfakeDetection

Zero-shot World Models via Search in Memory World Models have vastly permeated the field of Reinforcement Learning. Their ability to model the transition dynamics of an environment have greatly improved sample efficiency in online RL. Among the...

Our paper “Zero-shot World Models via Search in Memory” (by F. Malato & @villeh.bsky.social) was accepted to #NeurIPS2025! 🎉 A training-free world model predicting dynamics via memory search.

Poster: Exhibit Hall C,D,E on Wed 3 Dec 4:30–7:30 PM PST 🔗 arxiv.org/abs/2510.16123

See you in San Diego!

Hello BlueSky! We're StatML, the Statistical Machine Learning research group at the University of Eastern Finland. We study AI and Reinforcement Learning from multiple perspectives. Our website is launching soon, and we can’t wait to share more about our work!