Search
Search results for: continual learning
-
MagMax: Leveraging Model Merging for Seamless Continual Learning
PublicationThis paper introduces a continual learning approach named MagMax, which utilizes model merging to enable large pre-trained models to continuously learn from new data without forgetting previously acquired knowledge. Distinct from traditional continual learning methods that aim to reduce forgetting during task training, MagMax combines sequential fine-tuning with a maximum magnitude weight selection for effective knowledge integration...
-
Divide and not forget: Ensemble of selectively trained experts in Continual Learning
PublicationClass-incremental learning is becoming more popular as it helps models widen their applicability while not forgetting what they already know. A trend in this area is to use a mixture-of-expert technique, where different models work together to solve the task. However, the experts are usually trained all at once using whole task data, which makes them all prone to forgetting and increasing computational burden. To address this limitation,...
-
Looking through the past: better knowledge retention for generative replay in continual learning
PublicationIn this work, we improve the generative replay in a continual learning setting to perform well on challenging scenarios. Because of the growing complexity of continual learning tasks, it is becoming more popular, to apply the generative replay technique in the feature space instead of image space. Nevertheless, such an approach does not come without limitations. In particular, we notice the degradation of the continually trained...
-
Adapt Your Teacher: Improving Knowledge Distillation for Exemplar-free Continual Learning
PublicationIn this work, we investigate exemplar-free class incremental learning (CIL) with knowledge distillation (KD) as a regularization strategy, aiming to prevent forgetting. KDbased methods are successfully used in CIL, but they often struggle to regularize the model without access to exemplars of the training data from previous tasks. Our analysis reveals that this issue originates from substantial representation shifts in the teacher...
-
Revisiting Supervision for Continual Representation Learning
Publication"In the field of continual learning, models are designed to learn tasks one after the other. While most research has centered on supervised continual learning, there is a growing interest in unsupervised continual learning, which makes use of the vast amounts of unlabeled data. Recent studies have highlighted the strengths of unsupervised methods, particularly self-supervised learning, in providing robust representations. The improved...
-
Category Adaptation Meets Projected Distillation in Generalized Continual Category Discovery
Publication"Generalized Continual Category Discovery (GCCD) tackles learning from sequentially arriving, partially labeled datasets while uncovering new categories. Traditional methods depend on feature distillation to prevent forgetting the old knowledge. However, this strategy restricts the model’s ability to adapt and effectively distinguish new categories. To address this, we introduce a novel technique integrating a learnable projector...
-
Vident-synth: a synthetic intra-oral video dataset for optical flow estimation
Open Research DataWe introduce Vident-synth, a large dataset of synthetic dental videos with corresponding ground truth forward and backward optical flows and occlusion masks. It can be used for:
-
A Note on Reduced Strain Gradient Elasticity
PublicationWe discuss the particular class of strain-gradient elastic material models which we called the reduced or degenerated strain-gradient elasticity. For this class the strain energy density depends on functions which have different differential properties in different spatial directions. As an example of such media we consider the continual models of pantographic beam lattices and smectic and columnar liquid crystals.
-
A Nonlinear Model of a Mesh Shell
PublicationFor a certain class of elastic lattice shells experiencing finite deformations, a continual model using the equations of the so-called six-parameter shell theory has been proposed. Within this model, the kinematics of the shell is described using six kinematically independent scalar degrees of freedom — the field of displacements and turns, as in the case of the Cosserat continuum, which gives reason to call the model under consideration...
-
On the Equations of the Surface Elasticity Model Based on the Theory of Polymeric Brushes
PublicationMotivating by theory of polymers, in particular, by the models of polymeric brushes we present here the homogenized (continual) two-dimensional (2D) model of surface elasticity. A polymeric brush consists of an system of almost aligned rigid polymeric chains. The interaction between chain links are described through Stockmayer potential, which take into account also dipole-dipole interactions. The presented 2D model can be treated...
-
Prototype of a low-cost luxmeter with wide measuring range designed for railway stations dynamic lighting systems
PublicationThe paper presents the development of a low-cost luxmeter designed for usage in automatic applications for continual (long-term) measurement of illuminance. The measured information is used to adjust the level of lighting systems of artificial lighting on railways (platforms, marshalling yard, tracks etc.). The reason of this measuring device is a requirement for massive deployment of dimmable lighting systems for railways, in...
-
The influence of applying supplementary web plates on structural joint moment resistance and rotational stiffness
PublicationContinual trend to simplify structural elements and find less time-consuming solutions on construction site as well as inevitable increase in labour costs determine for instance elimination of those technological steps that are difficult to automation. Welding additional reinforcements and web stiffeners is a great example of such a step. The method presented in EN 1993-1-8 [8] is a result of long-term studies focused on calculating...
-
Analysis of positioning methods using Global Navigation Satellite Systems (GNSS) in Polish State Railways (PKP)
PublicationEach year, global navigation satellite systems (GNSS) improve their accuracy, availability, continuity, integrity, and reliability. Due to these continual improvements, the systems are increasingly used in various modes of transport, including rail transport, the subject of this publication. GNSS are used for rail passenger information, rail traffic management, and rail traffic control. These applications differ in the positioning...
-
Hybrid quantum-classical approach for atomistic simulation of metallic systems
PublicationThe learn-on-the-fly (LOTF) method [G. Csanyi et al., Phys. Rev. Lett. 93, 175503 (2004)] serves to seamlessly embed quantum-mechanical computations within a molecular-dynamics framework by continual local retuning of the potential's parameters so that it reproduces the quantum-mechanical forces. In its current formulation, it is suitable for systems where the interaction is short-ranged, such as covalently bonded semiconductors....
-
Homogeneity study of candidate reference material (contaminated soil) based on determination of selected metals, PCBs and PAHs
PublicationReference materials (RMs) are an indispensable part of QA/QC system and have played a critical role in the continual expansion of analytical capabilities of the analytical laboratories. The homogeneity study of the candidate for reference material is essential for the quality of reference material (certified reference material) used in quality assurance activities. The homogeneity study of candidate for reference materials of contaminated...
-
Looking Up to the Stars. A Call for Action to Save New Zealand’s Dark Skies for Future Generations to Come
PublicationThe rapid development of technology coupled with humanity’s desire to reach beyond terra firma, has resulted in more than 60 years of Outer Space activities. Although the exploration of space has provided many advantages and benefits to society so far, including vast, new information that has greatly added to our understanding of our planet and beyond, unfortunately, mankind’s footprint has negative aspects that need to be minimised...
-
Neoclassical Navier–Stokes Equations Considering the Gyftopoulos–Beretta Exposition of Thermodynamics
PublicationThe seminal Navier-Stokes equations were stated even before the creation of the foundations of thermodynamics and its first and second laws. There is a widespread opinion in the literature on thermodynamic cycles that the Navier-Stokes equations cannot be taken as a thermodynamically correct model of a local "working fluid", which would be able to describe the conversion of "heating" into "working" (Carnot's type cycles) and vice...
-
Piotr Szczuko dr hab. inż.
PeoplePiotr Szczuko received his M.Sc. degree in 2002. His thesis was dedicated to examination of correlation phenomena between perception of sound and vision for surround sound and digital image. He finished Ph.D. studies in 2007 and one year later completed a dissertation "Application of Fuzzy Rules in Computer Character Animation" that received award of Prime Minister of Poland. His interests include: processing of audio and video, computer...
-
Antitumor 1-nitroacridine derivative C-1748 induces significant apoptosis in pancreatic cancer cells.
PublicationPancreatic cancer is the fifth leading cause of cancer death and has the lowest survival rate of any solid cancer in the industrial countries. The poor prognosis of pancreatic cancer results from its tendency for late presentation, aggressive invasion, early metastasis, and resistance to chemotherapy. Gemcitabine still remains the best chemotherapeutic agent available for the treatment of advanced pancreatic cancer. However, gemcitabine...