Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Seemingly Redundant Modules Enhance Robust Odor Learning in Fruit Flies
Authors: HaiYang Li, Liao Yu, Qiang Yu, Yunliang Zang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this study, we present a computational model of the fly olfactory circuit to investigate odor discrimination under varying noise conditions that simulate complex environments. Our results show that LI primarily enhances odor discrimination in lowand medium-noise scenarios, but this benefit diminishes and may reverse under higher-noise conditions. In contrast, SFA consistently improves discrimination across all noise levels. ... We developed a spiking neural network model of the fly olfactory circuit. ... 3 Experiments ... Table 1 presents a comparative analysis of discrimination performance under varying noise intensities... |
| Researcher Affiliation | Academia | Haiyang Li1* Liao Yu2* Qiang Yu3 Yunliang Zang1,4 1Academy of Medical Engineering and Translational Medicine, Tianjin University, Tianjin, China 2School of Mathematical Sciences, Beihang University, Beijing, China 3School of Artificial Intelligence, Tianjin University, Tianjin, China 4Xiamen Intretech Inc, Xiamen, Fujian, China |
| Pseudocode | Yes | Algorithm 1 Simplified SNN Training with Adaptive Mechanisms for Olfactory Tasks |
| Open Source Code | Yes | The code is available at: https://github.com/L-0cean/Fly-SNN. |
| Open Datasets | Yes | Input Data. We adapted the odor-discrimination dataset from [43]. For each odor class i, we define a prototype ORN response vector Ii R50. Its components are independently sampled from a uniform distribution: Ii,j U(0, 1) for j = 1, . . . , 50. An individual sample from class i, denoted I(k) i , is generated by adding an additive noise vector to the prototype: I(k) i = Ii + Y(k), where every component of the noise vector Y(k) R50 is independently sampled from a zero-mean Gaussian distribution, Y (k) j N(0, σ2 noise). In accordance with the non-negativity of ORN firing rates, we applied element-wise clipping so that I(k) i 0. [43] Peter Y Wang, Yi Sun, Richard Axel, LF Abbott, and Guangyu Robert Yang. Evolving the olfactory system with machine learning. Neuron, 109(23):3879 3892, 2021. |
| Dataset Splits | Yes | Dataset and models: We generated an odor dataset comprising 30,000 training samples and 10,000 test samples, each drawn at random as described above. |
| Hardware Specification | Yes | Simulations were implemented in Python (snn Torch) and run on an NVIDIA A800 GPU. |
| Software Dependencies | No | Simulations were implemented in Python (snn Torch) and run on an NVIDIA A800 GPU. |
| Experiment Setup | Yes | Experimental settings: Simulations were implemented in Python (snn Torch) and run on an NVIDIA A800 GPU. Network dynamics were simulated with a 1-ms time step; each trial comprised a 10-ms pre-stimulus baseline followed by a 30-ms odor presentation. All neurons shared a membrane time constant of 10 ms. Firing thresholds were 0.8 for PNs, LNs, and KCs, and 1.2 for MBONs. Connectivity featured sparse PN KC projections (each KC received inputs from 6 PNs; fixed weight 0.3) and fully connected KC MBON synapses initialized uniformly in [0, 0.08]. LI used a 5-ms LN trace time constant, and SFA used a 50-ms adaptation time constant. Models were trained for 100 epochs (batch size 256) with Adam (initial learning rate 1.0 10 4). |