Lee Cooper

I'm glad to share our new paper, which reports on the development of an e-tongue for sensitive monitoring of neurotransmitters using four different #MXenes as sensing units. Note that the first two authors were visiting scientists from Brazil and Azerbaijan at A.J. Drexel Nanomaterials Institute. Without them, this work would not be done, and we would not even start it. This shows the importance of bringing creative international visitors who implement their original ideas. Increasing restrictions will badly hurt the scientific productivity of US research labs. #research #science #MXene, #etongue #sensors #collaboration #visitingscientists

166

7 Comments

I'm thrilled to announce that my latest research paper on melanoma detection has been published in SAGE Journals! (Farrukh A. Khan, PhD, FBCS, SMIEEE, Mamoona HUmayun) Discover how our approach enhances privacy and accuracy in medical imaging. Read it here: https://lnkd.in/eckYtskQ I look forward to your thoughts and feedback! #Research #MelanomaDetection #HyperParameterTuning #CNN #MedicalImaging #AcademicPublishing

21

7 Comments

I recently read “Towards Monosemanticity: Decomposing Language Models with Dictionary Learning” by Bricken et al 2023. Published at transformer-circuits.pub In this paper the authors discuss the use of autoencoders to extract interpretable features from a one-layer transformer model. The approach aims to improve the understanding and transparency of artificial neural networks. To do this they use the concept of dictionary learning which is a method used to decompose the activations of the transformer model into a set of basis features that are sparse and interpretable. This involves training sparse autoencoders to identify a dictionary of monosemantic features (or features with a single meaning). These monosemantic features in single units (artificial neurons) reminds me of the “grandmother neuron” theory in neuroscience which proposes that specific neurons respond to specific highly complex stimuli. However, in real brains, it is widely believed that complex representations are encoded by the collective activity of groups of neurons (and not monosemantic). But going back to the paper, I really like the visualization tools and illustration to make such complex networks more transparent (as opposed to black boxes). Paper: https://lnkd.in/gR87pJcs

3

The Veridical Data Science talks at the Sutardja Dai Hall of UC Berkeley was quite informative about reproducibility, replication and publication bias. Prof. Bin Yu outlined her modeling principles based on PCS: Predictability, Computability and Stability. I myself had started thinking about borrowing the stability computations from control system theory; so it was reaffirming to hear that I was on the right track. We took home lots'a marks from the lively discussions. Dr. Yu's book from MiT Press is freely available through vdsbook.com, and software at https://yu-group-github.

19

1 Comment

I got a new textbook on Melrose's scattering calculus: Richard B. Melrose Spectral and scattering theory for the Laplacian on asymptotically Euclidian spaces in Scattering Theory Mitsuru Ikawa eds CRC press, 1994. https://lnkd.in/gHzqKzj7

1

This weekend I read the paper "Advancing Multimodal Medical Capabilities of Gemini" by a group from Google Research and Google DeepMind. In this paper the authors show their results after fine-tuning Gemini, a foundation multimodal model for a variety of medical tasks. The authors use combinations of open and private 2D and 3D radiology, histopathology, ophthalmology, dermatology and genomic datasets. The tasks for which this decoding transformer model was fine-tuned included Question answering, Report generation, genomic disease prediction and different condition-associated image classification. To me, this paper was interesting because it shows the amount of data required to fine-tune such big models. I am interested also in resources that compile open medical datasets that can be used to train multimodal models (especially in dental imagery). On this regard, the authors briefly describe how the medical specialized gemini (Med-Gemini), was trained on large-scale Google TPUv4 accelerator pods spread across multiple data-centers. I wish more details were given about costs, duration and data engineering for the training and evaluation tasks (although they provide links to other projects from Google Research). Paper: https://lnkd.in/gjjpiCSw #gemini #LMMs #medAI

28

1 Comment

A student recently asked me if I had a favorite #neuroscience paper. While there are many influential studies I could list, this one from Tiago Branco's group is probably my favorite (link below). I love it because the authors elegantly show how computations within a simple neural circuit rapidly control an evolutionarily conserved defensive behavior in mice (escape from an aerial predator). Evans et al.'s findings guide my lab's studies of #neurofibromatosis, because we've found that mice modeling #NF1 and other #Rasopathy syndromes have exaggerated responses to looming discs that mimic predator approach from above. We think these behaviors could mimic sensory hypersensitivity experienced by some NF patients with autism and may involve evolutionarily old brain systems that control emotional responses to sensory information. I'll post more about our findings in the near future! #EndNF

13

1 Comment

Important step towards improving transparency, reproducibility, and scalability of the AI analysis tools in biomedical imaging data. AI-generated segmentations are now available along with data in the National Lung Screening Trial (NLST) collection. The results were generated by applying Totalsegmentator in cloud to 126,088 CT series to segment the total of 9,565,554 anatomic structures.

34

1 Comment

I'm thrilled to share that our recent paper titled "BERT-Based Deceptive Review Detection in Social Media: Introducing DeceptiveBERT" has been published in the IEEE Transactions on Computational Social Systems and is now available in the "Early Access" section on IEEE Xplore. This work introduces DeceptiveBERT, a novel methodology utilizing BERT for the detection of deceptive reviews on social media platforms. It's a significant step forward in understanding and mitigating the impacts of misinformation in digital communications. You can read the full abstract and access the early version of the paper here: https://lnkd.in/esSdhrr2 DOI: 10.1109/TCSS.2024.3403937 #SocialMedia #MachineLearning #DeceptiveReviewDetection #IEEE #ResearchImpact

16

4 Comments

Conditional Computation in LLMs (MoEs): Understanding the Concept of MoEs The Mixture of Experts (MoE) works on the idea of conditional computation, allowing the model to only run through some parts (an Expert) for a given input compared to a dense model where all parameters are used. It consists of a gating mechanism that selects the experts for the given input, enabling the model to scale without increasing the required computation. Popular LLM architectures like Mixtral and Grok use the same concept. What is MoEs? The Mixture of Experts (MoE) makes a simple modification to the usual architecture by replacing all or some of the Feed Forward layers with an MoE layer. The MoE layer consists of: 1. Router: A parameter gating mechanism that selects the experts used to process the given input by producing a probability distribution over experts. 2. Experts: Standalone NN modules with independent sets of parameters trained jointly with routers. It can use more complex architecture to create hierarchical MoE modules by implementing each expert as another MoE. - Implementation in Switch Transformers : Training was found to be 7x faster than training the base T5, keeping computation and training time constant. - Gshard and MeshTransformer : These frameworks propose efficient methods to train the MoE architectures using the parallelisation of models across GPUs/TPUs. Challenges with MoEs and Solutions 1. Inconsistent Batch Size for Experts: Can be overcome with the use of "soft" constraints in training loss. 2. Load Balancing Across Experts: The issue of repeatedly using the same few experts while training can be overcome by including better balancing mechanisms. 3. Model Instability Issues: Use of router-z Loss as an auxiliary loss to encourage smaller values within the router mechanism. 4. Model Overfitting: Requires different hyperparameter settings for effective fine-tuning. It was found that models benefit from high learning rates and smaller batch sizes. Recommended Reading/References 1. Hugging Face Blog on MoE - https://lnkd.in/g36gQJTx 2. Cameron Wolfe's Substack on Conditional Computation - https://lnkd.in/gtpjMHnJ

60

1 Comment

🎉 Exciting News! 🎉 I am happy to share that my latest paper, titled "Degradation assessment of an IGBT with recurrence analysis and Kalman filter-based data fusion" has been published in Chaos, Solitons, & Fractals! This publication is the first output of our TUBITAK project, "Integrated Condition Assessment through Kalman Filter-based Data Fusion and Recurrence Analysis-Aided Aging Monitoring." This paper presents a novel tool for assessing IGBT degradation by using recurrence analysis and Kalman filter-based data fusion to effectively handle nonstationary processes. For the next 50 days, the article is freely accessible! I invite all my connections to read and share your thoughts on this innovative approach. Read the article here: https://lnkd.in/d_hDXZdJ Thank you for your support and engagement! #ICMDLab #ITU #DegradationAssessment #RecurrenceAnalysis #KalmanFilter #DataFusion #ChaosSolitonsFractals #IGBT #Agingmonitoring #TechnologyInnovation #ScienceDirect #EngineeringResearch

90

5 Comments

Our SCIE article titled Integrating fuzzy metrics and negation operator in FCM algorithm via genetic algorithm for MRI image segmentation has been accepted. ✌️ 🚀 https://lnkd.in/dEpBnfNM

20

1 Comment

Joint work with Hemant Mishra and Prof. Michael Nussbaum published in Letters in Mathematical Physics: https://lnkd.in/g4WbFCqW https://lnkd.in/gU_XKUVq Title: On the optimal error exponents for classical and quantum antidistinguishability Abstract: The concept of antidistinguishability of quantum states has been studied to investigate foundational questions in quantum mechanics. It is also called quantum state elimination, because the goal of such a protocol is to guess which state, among finitely many chosen at random, the system is not prepared in (that is, it can be thought of as the first step in a process of elimination). Antidistinguishability has been used to investigate the reality of quantum states, ruling out 𝜓-epistemic ontological models of quantum mechanics (Pusey et al. in Nat Phys 8(6):475–478, 2012). Thus, due to the established importance of antidistinguishability in quantum mechanics, exploring it further is warranted. In this paper, we provide a comprehensive study of the optimal error exponent—the rate at which the optimal error probability vanishes to zero asymptotically—for classical and quantum antidistinguishability. We derive an exact expression for the optimal error exponent in the classical case and show that it is given by the multivariate classical Chernoff divergence. Our work thus provides this divergence with a meaningful operational interpretation as the optimal error exponent for antidistinguishing a set of probability measures. For the quantum case, we provide several bounds on the optimal error exponent: a lower bound given by the best pairwise Chernoff divergence of the states, a single-letter semi-definite programming upper bound, and lower and upper bounds in terms of minimal and maximal multivariate quantum Chernoff divergences. It remains an open problem to obtain an explicit expression for the optimal error exponent for quantum antidistinguishability.

18

4 Comments