Large Language models have demonstrated excellent domain-specific question-answering capabilities when finetuned with a particular dataset of that specific domain. However, fine-tuning the models requires a significant amount of training time and a considerable amount of hardware. In this work, we propose CARE (Customer Assistance and Response Engine), a lightweight model made by fine-tuning Phi3.5-mini on very minimal hardware and data, designed to handle queries primarily across three domains: telecommunications support, medical support, and banking support. For telecommunications and banking, the chatbot addresses issues and problems faced by customers regularly in the above-mentioned domains. In the medical domain, CARE provides preliminary support by offering basic diagnoses and medical suggestions that a user might take before consulting a healthcare professional. Since CARE is built on Phi3.5-mini, it can be used even on mobile devices, increasing its usability. Our research also shows that CARE performs relatively well on various medical benchmarks, indicating that it can be used to make basic medical suggestions.

We prove an oracle inequality for generic regularized empirical risk minimization algorithms learning from \a -mixing processes. To illustrate this oracle inequality, we use it to derive learning rates for some learning methods including least squares SVMs. Since the proof of the oracle inequality uses recent localization ideas developed for independent and identically distributed (i.i.d.) processes, it turns out that these learning rates are close to the optimal rates known in the i.i.d.

A class of fast, supervised learning algorithms is presented. Inspired by Albus's CMAC model, the algorithms learn orders of magnitude more rapidly than typical imple(cid:173) mentations of back propagation, while often achieving comparable qualities of generalization. Furthermore, unlike most traditional function approximation methods, the algorithms are well suited for use in real time adaptive signal processing. Unlike simpler adaptive systems, such as linear predictive cod(cid:173) ing, the adaptive linear combiner, and the Kalman filter, the new algorithms are capable of efficiently capturing the structure of complicated non-linear systems. As an illustration, the algorithm is applied to the prediction of a chaotic timeseries.

A method for transforming performance evaluation signals distal both in space and time into proximal signals usable by supervised learning algo(cid:173) rithms, presented in [Jordan & Jacobs 90], is examined. A simple obser(cid:173) vation concerning differentiation through models trained with redundant inputs (as one of their networks is) explains a weakness in the original architecture and suggests a modification: an internal world model that encodes action-space exploration and, crucially, cancels input redundancy to the forward model is added. Learning time on an example task, cart(cid:173) pole balancing, is thereby reduced about 50 to 100 times.

Sigmoid type belief networks, a class of probabilistic neural net(cid:173) works, provide a natural framework for compactly representing probabilistic information in a variety of unsupervised and super(cid:173) vised learning problems. Often the parameters used in these net(cid:173) works need to be learned from examples. Unfortunately, estimat(cid:173) ing the parameters via exact probabilistic calculations (i.e, the EM-algorithm) is intractable even for networks with fairly small numbers of hidden units. We propose to avoid the infeasibility of the E step by bounding likelihoods instead of computing them ex(cid:173) actly. We introduce extended and complementary representations for these networks and show that the estimation of the network parameters can be made fast (reduced to quadratic optimization) by performing the estimation in either of the alternative domains.

We develop Few-Shot Learning models trained to recognize five or ten different dynamic hand gestures, respectively, which are arbitrarily interchangeable by providing the model with one, two, or five examples per hand gesture. All models were built in the Few-Shot Learning architecture of the Relation Network (RN), in which Long-Short-Term Memory cells form the backbone. The models use hand reference points extracted from RGB-video sequences of the Jester dataset which was modified to contain 190 different types of hand gestures. Result show accuracy of up to 88.8% for recognition of five and up to 81.2% for ten dynamic hand gestures. The research also sheds light on the potential effort savings of using a Few-Shot Learning approach instead of a traditional Deep Learning approach to detect dynamic hand gestures. Savings were defined as the number of additional observations required when a Deep Learning model is trained on new hand gestures instead of a Few Shot Learning model. The difference with respect to the total number of observations required to achieve approximately the same accuracy indicates potential savings of up to 630 observations for five and up to 1260 observations for ten hand gestures to be recognized. Since labeling video recordings of hand gestures implies significant effort, these savings can be considered substantial.

Pairwise learning corresponds to the supervised learning setting where the goal is to make predictions for pairs of objects. Prominent applications include predicting drug-target or protein-protein interactions, or customer-product preferences. Several kernel functions have been proposed for incorporating prior knowledge about the relationship between the objects, when training kernel based learning methods. However, the number of training pairs n is often very large, making O(n^2) cost of constructing the pairwise kernel matrix infeasible. If each training pair x= (d,t) consists of drug d and target t, let m and q denote the number of unique drugs and targets appearing in the training pairs. In many real-world applications m,q << n, which can be used to develop computational shortcuts. Recently, a O(nm+nq) time algorithm we refer to as the generalized vec trick was introduced for training kernel methods with the Kronecker kernel. In this work, we show that a large class of pairwise kernels can be expressed as a sum of product matrices, which generalizes the result to the most commonly used pairwise kernels. This includes symmetric and anti-symmetric, metric-learning, Cartesian, ranking, as well as linear, polynomial and Gaussian kernels. In the experiments, we demonstrate how the introduced approach allows scaling pairwise kernels to much larger data sets than previously feasible, and compare the kernels on a number of biological interaction prediction tasks.

Although graph neural networks (GNNs) have made great progress recently on learning from graph-structured data in practice, their theoretical guarantee on generalizability remains elusive in the literature. In this paper, we provide a theoretically-grounded generalizability analysis of GNNs with one hidden layer for both regression and binary classification problems. Under the assumption that there exists a ground-truth GNN model (with zero generalization error), the objective of GNN learning is to estimate the ground-truth GNN parameters from the training data. To achieve this objective, we propose a learning algorithm that is built on tensor initialization and accelerated gradient descent. We then show that the proposed learning algorithm converges to the ground-truth GNN model for the regression problem, and to a model sufficiently close to the ground-truth for the binary classification problem. Moreover, for both cases, the convergence rate of the proposed learning algorithm is proven to be linear and faster than the vanilla gradient descent algorithm. We further explore the relationship between the sample complexity of GNNs and their underlying graph properties. Lastly, we provide numerical experiments to demonstrate the validity of our analysis and the effectiveness of the proposed learning algorithm for GNNs.

We prove an oracle inequality for generic regularized empirical risk minimization algorithms learning from $\a$-mixing processes. To illustrate this oracle inequality, we use it to derive learning rates for some learning methods including least squares SVMs. Since the proof of the oracle inequality uses recent localization ideas developed for independent and identically distributed (i.i.d.) processes, it turns out that these learning rates are close to the optimal rates known in the i.i.d. Papers published at the Neural Information Processing Systems Conference.

Stochastic neighbor embedding (SNE) and related nonlinear manifold learning algorithms achieve high-quality low-dimensional representations of similarity data, but are notoriously slow to train. We propose a generic formulation of embedding algorithms that includes SNE and other existing algorithms, and study their relation with spectral methods and graph Laplacians. This allows us to define several partial-Hessian optimization strategies, characterize their global and local convergence, and evaluate them empirically. We achieve up to two orders of magnitude speedup over existing training methods with a strategy (which we call the spectral direction) that adds nearly no overhead to the gradient and yet is simple, scalable and applicable to several existing and future embedding algorithms.