Journal Club

Today's article comes from the journal of Machine Learning and Knowledge Extraction. The authors are Maltoni et al., from the University of Bologna, in Italy. In this paper, They want to know two things: How a transformer actually implements multi-step deductive reasoning internally. And whether that computation differs when the reasoning happens through generated intermediate tokens versus through the model's layers.

Today's article comes from the journal of Memories - Materials, Devices, Circuits and Systems. The authors are Shih et al., from Imec, in Belgium. In this paper they're showcasing a new type of computer memory. It works by storing binary information as ordered chains of two different nanoparticle types inside billions of liquid-filled nanocapillaries.

Today's article comes from the journal of Data-Centric Engineering. The authors are Savasta et al., from the Olesinki firm, in the United Kingdom. In this paper they're taking a new algorithmic approach to motorboat design. Whereas a normal process might need to solve for geometric fit, adjacency, or accessibility, these designers also need to consider something that is much trickier to codify: visual aesthetics.

Today's article comes from the journal of Machine Learning: Science and Technology. The authors are Harb et al., from Argonne National Laboratory, in Illinois. In this paper, they're showcasing a new prompt-level reasoning enhancement framework, that they claim can improve LLM performance on ChemBench by several percentage points.

Today's article comes from the journal of Materials Genome Engineering Advances. The authors are Shu et al., from Xi'an Jiaotong University, in China. In this paper, they've taken DeepSeek-R1 and fine-tuned it for a very specific task: lattice-structure design.

Today's article comes from the SoftwareX journal. The authors are Rodriguez-Cardos et al., from the University of Castilla-La Mancha, in Spain. In this paper, they're showcasing a new experimentation platform for Elastic Patterns research.

Today's article comes from the Journal of Medical Internet Research. The authors are Kumar et al., from the University of New Mexico Health Sciences Center. In this paper they're putting forward a solution for medical "undercoding" that uses a relatively specialized form of machine learning: Positive Unlabeled (PU).

Today's article comes from the BMC Bioinformatics journal. The authors are Stadelmaier et al., from the Eberhard Karls University of Tubingen. In this paper the authors build several models for t-cell response prediction. But, they also create an an evaluation scheme that allows them to detect when their models' performance is being inflated by shortcut learning.

Today's article comes from the Advances in Human-Computer Interaction journal. The authors are AbuSafya et al., from Al-Ahliyya Amman University, in Jordan. In this paper they're trying to make the B-tree data structure a little more developer-friendly, a little easier to adapt, and a little more transparent.

Today's article comes from the journal of Computer Assisted Surgery. The authors are Connolly et al., from Queen's University in Canada. In this paper they're showcasing an experimental testbed for photoacoustic imaging. It's set up to let researchers prototype, automate, and compare different ways of scanning a tumor bed.

Today's article comes from the journal of Machine Learning: Science and Technology. The authors are Chen et al., from Pacific Northwest National Laboratory (PNNL), in Washington. In this paper they make the argument that many PINN failures are simultaneously optimization problems and sampling problems. Their solution? Instead of treating adaptive weighting and adaptive sampling as competing alternatives, just combine them into a single unified framework.

Today's article comes from the MGE Advances journal. The authors are Nie et al., from Northwestern Polytechnical University, in China. In this paper they're showcasing a pipeline that can hunt for new metals with specific physical properties. It screens through a massive set of candidate compositions, and narrows those down to a handful of promising alloys.

Today's article comes from the journal Array. The authors are Min et al., from Ewha Womans University, in South Korea. In this paper, they present a framework that helps you choose the right learning rate for every layer in a model.

Today's article comes from the SoftwareX journal. The authors are Lattanzi et al., from the Institute of Marine Biological Resources and Biotechnologies, in Italy. In this paper they're building a binary classifier that can look at the GPS trajectory of a boat and figure out whether its fishing or not.

Today's article comes from the Cambridge Forum on AI: Law and Governance. The authors are Basu et al., from the University of Lille, in France. In this paper, they're proposing a bias-reduction framework called "Fair Game". It wraps a model inside a loop, between an auditor and a debiasing algorithm. And the whole interaction gets structured as a reinforcement learning problem.

Today's article comes from the journal of Advanced Intelligent Systems. The authors are Le et al., from the University of Queensland, in Australia. This paper is an exploration of what's possible when you try to turn insects (like cockroaches, beetles, and grasshoppers), into remote-controlled cyborgs.

Today's article comes from the JASA Express Letters journal. The authors are Ragland et al., from Woods Hole Oceanographic Institution, in Massachusetts. In this paper, the authors take data collected from a set of OOI's hydrophones, and use it to figure out how much low-frequency underwater-sound is controlled by wind speed.

Today's article comes from the journal of Big Data. The authors are Emam et al., from Minia University, in Egypt. In this paper, they explore a new variation of a popular feature-selection algorithm.

Today's article comes from the International Journal of Computational Intelligence Systems. The authors are Chilukuri et al., from St. Jude Childrens Cancer Research Hospital, in Tennessee. In this paper they're proposing a two-stage deep learning system for group recommendations.

Today's article comes from the Frontiers in Computer Science journal. The authors are Xiao et al., from Hunan Institute of Engineering, in China. In this paper, they're taking the signals that fault localization normally uses and augmenting them with static features, derived from the repo.

Today's article comes from the journal of Machine Learning and Knowledge Extraction. The authors are Nassar et al., from Hamad Bin Khalifa University, in Qatar. In this paper they're evaluating two replacements for KL-divergence within t-SNE. Max-Flipped KL Divergence (KLmax) and KL-Wasserstein Loss.

Today's article comes from the Advances in Fuzzy Systems journal. The authors are Zhang et al., from Anqing Normal University, in China. In this paper they're exploring a new strategy for solving MCDMs with fuzzy inputs. Their approach works by representing uncertain evaluations as sets of possible values, measuring the distance between those sets without distorting the data, and then deriving the importance of each criterion from the level of disagreement it creates among the alternatives.

Today's article comes from the journal of Machine Learning Science & Technology. The authors are Nikolaou et al., from the University of Stuttgart, in Germany. In this paper they use NTK (the Neural Tangent Kernel) as a lens to study what actually happens inside neural networks as we scale them up.