Continual lifelong learning with neural networks: A review.

Humans and animals have the ability to continually acquire, fine-tune, and transfer knowledge and skills throughout their lifespan. This ability, referred to as lifelong learning, is mediated by a rich set of neurocognitive mechanisms that together contribute to the development and specialization of our sensorimotor skills as well as to long-term memory consolidation and retrieval. Consequently, lifelong learning capabilities are crucial for computational learning systems and autonomous agents interacting in the real world and processing continuous streams of information. However, lifelong learning remains a long-standing challenge for machine learning and neural network models since the continual acquisition of incrementally available information from non-stationary data distributions generally leads to catastrophic forgetting or interference. This limitation represents a major drawback for state-of-the-art deep neural network models that typically learn representations from stationary batches of training data, thus without accounting for situations in which information becomes incrementally available over time. In this review, we critically summarize the main challenges linked to lifelong learning for artificial learning systems and compare existing neural network approaches that alleviate, to different extents, catastrophic forgetting. Although significant advances have been made in domain-specific learning with neural networks, extensive research efforts are required for the development of robust lifelong learning on autonomous agents and robots. We discuss well-established and emerging research motivated by lifelong learning factors in biological systems such as structural plasticity, memory replay, curriculum and transfer learning, intrinsic motivation, and multisensory integration.

Compositional Learning of Human Activities With a Self-Organizing Neural Architecture.

An important step for assistive systems and robot companions operating in human environments is to learn the compositionality of human activities, i.e., recognize both activities and their comprising actions. Most existing approaches address action and activity recognition as separate tasks, i.e., actions need to be inferred before the activity labels, and are thus highly sensitive to the correct temporal segmentation of the activity sequences. In this paper, we present a novel learning approach that jointly learns human activities on two levels of semantic and temporal complexity: (1) transitive actions such as reaching and opening, e.g., a cereal box, and (2) high-level activities such as having breakfast. Our model consists of a hierarchy of GWR networks which process and learn inherent spatiotemporal dependencies of multiple visual cues extracted from the human body skeletal representation and the interaction with objects. The neural architecture learns and semantically segments input RGB-D sequences of high-level activities into their composing actions, without supervision. We investigate the performance of our architecture with a set of experiments on a publicly available benchmark dataset. The experimental results show that our approach outperforms the state of the art with respect to the classification of the high-level activities. Additionally, we introduce a novel top-down modulation mechanism to the architecture which uses the actions and activity labels as constraints during the learning phase. In our experiments, we show how this mechanism can be used to control the network's neural growth without decreasing the overall performance.

Expectation Learning for Stimulus Prediction Across Modalities Improves Unisensory Classification.

Expectation learning is a unsupervised learning process which uses multisensory bindings to enhance unisensory perception. For instance, as humans, we learn to associate a barking sound with the visual appearance of a dog, and we continuously fine-tune this association over time, as we learn, e.g., to associate high-pitched barking with small dogs. In this work, we address the problem of developing a computational model that addresses important properties of expectation learning, in particular focusing on the lack of explicit external supervision other than temporal co-occurrence. To this end, we present a novel hybrid neural model based on audio-visual autoencoders and a recurrent self-organizing network for multisensory bindings that facilitate stimulus reconstructions across different sensory modalities. We refer to this mechanism as stimulus prediction across modalities and demonstrate that the proposed model is capable of learning concept bindings by evaluating it on unisensory classification tasks for audio-visual stimuli using the 43,500 Youtube videos from the animal subset of the AudioSet corpus.

Lifelong Learning of Spatiotemporal Representations With Dual-Memory Recurrent Self-Organization.

Artificial autonomous agents and robots interacting in complex environments are required to continually acquire and fine-tune knowledge over sustained periods of time. The ability to learn from continuous streams of information is referred to as lifelong learning and represents a long-standing challenge for neural network models due to catastrophic forgetting in which novel sensory experience interferes with existing representations and leads to abrupt decreases in the performance on previously acquired knowledge. Computational models of lifelong learning typically alleviate catastrophic forgetting in experimental scenarios with given datasets of static images and limited complexity, thereby differing significantly from the conditions artificial agents are exposed to. In more natural settings, sequential information may become progressively available over time and access to previous experience may be restricted. Therefore, specialized neural network mechanisms are required that adapt to novel sequential experience while preventing disruptive interference with existing representations. In this paper, we propose a dual-memory self-organizing architecture for lifelong learning scenarios. The architecture comprises two growing recurrent networks with the complementary tasks of learning object instances (episodic memory) and categories (semantic memory). Both growing networks can expand in response to novel sensory experience: the episodic memory learns fine-grained spatiotemporal representations of object instances in an unsupervised fashion while the semantic memory uses task-relevant signals to regulate structural plasticity levels and develop more compact representations from episodic experience. For the consolidation of knowledge in the absence of external sensory input, the episodic memory periodically replays trajectories of neural reactivations. We evaluate the proposed model on the CORe50 benchmark dataset for continuous object recognition, showing that we significantly outperform current methods of lifelong learning in three different incremental learning scenarios.

Inducing hardening and healability in poly(ethylene-co-acrylic acid) via blending with complementary low molecular weight additives.

The design and synthesis of low molecular weight additives based on self-assembling nitroarylurea units, and their compatibility with poly(ethylene-co-acrylic acid) copolymers are reported. The self-assembly properties of the low molecular weight additives have been demonstrated in a series of gelation studies. Upon blending at low percentage weights (≤5%) with poly(ethylene-co-acrylic acid) the additives were capable of increasing the stress and strain to failure when compared to the parent copolymer. By varying the percentage weight of the additive as well as the type of additive the mechanical properties of poly(ethylene-co-acrylic acid) could be tailored. Finally, the healability characteristics of the blends were improved when compared to the original polymer via the introduction of a supramolecular 'network within a network'.

Anti-Candida Activity of Bursera morelensis Ramirez Essential Oil and Two Compounds, α-Pinene and γ-Terpinene-An In Vitro Study.

The candidiasis caused by C. albicans is a public health problem. The abuse of antifungals has contributed to the development of resistance. B. morelensis has demonstrated antibacterial and antifungal activities. In this work the activity of the essential oil of B. morelensis was evaluated and for its two pure compounds with analysis of the different mechanisms of pathogenesis important for C. albicans. The essential oil was obtained by the hydro-distillation method and analyzed using GC-MS. The anti-Candida activity was compared between to essential oil, α-Pinene and γ-Terpinene. GC-MS of the essential oil demonstrated the presence of 13 compounds. The essential oil showed antifungal activity against four C. albicans strains. The most sensitive strain was C. albicans 14065 (MFC 2.0 mg/mL and MIC50 0.125 mg/mL) with α-Pinene and γ-Terpinene having MFCs of 4.0 and 16.0 mg/mL respectively. The essential oil inhibited the growth of the germ tube in 87.94% (8.0 mg/mL). Furthermore, it was observed that the essential oil diminishes the transcription of the gene INT1. This work provides evidence that confirms the anti-Candida activity of the B. morelensis essential oil and its effect on the growth of the germ tube and transcription of the gene INT1.

Lifelong learning of human actions with deep neural network self-organization.

Lifelong learning is fundamental in autonomous robotics for the acquisition and fine-tuning of knowledge through experience. However, conventional deep neural models for action recognition from videos do not account for lifelong learning but rather learn a batch of training data with a predefined number of action classes and samples. Thus, there is the need to develop learning systems with the ability to incrementally process available perceptual cues and to adapt their responses over time. We propose a self-organizing neural architecture for incrementally learning to classify human actions from video sequences. The architecture comprises growing self-organizing networks equipped with recurrent neurons for processing time-varying patterns. We use a set of hierarchically arranged recurrent networks for the unsupervised learning of action representations with increasingly large spatiotemporal receptive fields. Lifelong learning is achieved in terms of prediction-driven neural dynamics in which the growth and the adaptation of the recurrent networks are driven by their capability to reconstruct temporally ordered input sequences. Experimental results on a classification task using two action benchmark datasets show that our model is competitive with state-of-the-art methods for batch learning also when a significant number of sample labels are missing or corrupted during training sessions. Additional experiments show the ability of our model to adapt to non-stationary input avoiding catastrophic interference.

Two new Liolaemus lizards from the Andean highlands of Southern Chile (Squamata, Iguania, Liolaemidae).

Liolaemus is a diverse genus of lizards, subdivided into two subgenera: Liolaemus (sensu stricto) and Eulaemus, distributed mainly in Chile and Argentina. The Liolaemus elongatus-kriegi complex is the most diverse group within Liolaemus (sensu stricto), especially the species closely related to Liolaemus elongatus, which form a clade currently comprising nine species. Several Chilean species of this group have been recently described, mainly from volcanoes and poorly explored mountains. Here molecular and morphological evidence are provided for a new species of the Liolaemus elongatus clade, which is characterized by its small size and lack of dorsal pattern, unusual features for the species of this group of lizards. Additionally, the lack of precloacal pores in males of Liolaemus (sensu stricto) is a trait found in few species, which do not constitute a monophyletic group. A second new southern Chilean species is also described, without precloacal pores and supported by molecular phylogenetics to be related to Liolaemus villaricensis. Both new species were found in the same locality, near a lake located in a pre-Andean zone with Araucaria and Nothofagus forest. The two species are dedicated to prominent Lonkos (tribal chiefs) of the Mapuche and Pehuenche people: Janequeo and Leftraru. Additionally, the phylogenetic results suggest that Liolaemus lonquimayensis is a synonym of Liolaemus elongatus.

Non-cytotoxic copper overload boosts mitochondrial energy metabolism to modulate cell proliferation and differentiation in the human erythroleukemic cell line K562.

Copper is integral to the mitochondrial respiratory complex IV and contributes to proliferation and differentiation, metabolic reprogramming and mitochondrial function. The K562 cell line was exposed to a non-cytotoxic copper overload to evaluate mitochondrial dynamics, function and cell fate. This induced higher rates of mitochondrial turnover given by an increase in mitochondrial fusion and fission events and in the autophagic flux. The appearance of smaller and condensed mitochondria was also observed. Bioenergetics activity included more respiratory complexes, higher oxygen consumption rate, superoxide production and ATP synthesis, with no decrease in membrane potential. Increased cell proliferation and inhibited differentiation also occurred. Non-cytotoxic copper levels can modify mitochondrial metabolism and cell fate, which could be used in cancer biology and regenerative medicine.

A new species of Liolaemus related to L. nigroviridis from the Andean highlands of Central Chile (Iguania, Liolaemidae).

The Liolaemus nigroviridis group is a clade of highland lizards endemic to Chile. These species are distributed from northern to central Chile, and currently there are no cases of sympatric distribution. This study describes a new species, Liolaemus uniformis sp. n., from this group, and provides a detailed morphological characterization and mitochondrial phylogeny using cytochrome-b. Liolaemus uniformis was found in sympatry with Liolaemus nigroviridis but noticeably differed in size, scalation, and markedly in the color pattern, without sexual dichromatism. This new species has probably been confused with Liolaemus monticola and Liolaemus bellii, both of which do not belong to the nigroviridis group. The taxonomic issues of this group that remain uncertain are also discussed.

Physicochemical property profile for brain permeability: comparative study by different approaches.

A comparative study of classification models of brain penetration by different approaches was carried out on a training set of 1000 chemicals and drugs, and an external test set of 100 drugs. Ten approaches were applied in this work: seven medicinal chemistry approaches (including "rule of 5" and multiparameter optimization) and also three SAR techniques: logistic regression (LR), random forest (RF) and support vector machine (SVM). Forty-one different medicinal chemistry descriptors representing diverse physicochemical properties were used in this work. Medicinal chemistry approaches based on the intuitive estimation of preference zones of CNS or non-CNS chemicals, with different rules and scoring functions, yield unbalanced models with poor classification accuracy. RF and SVM methods yielded 82% and 84% classification accuracy respectively for the external test set. LR was also successful in CNS/non-CNS (denoted in this study as CNS+/CNS-) classification and yielded an overall accuracy equivalent to that of SVM and RF. At the same time, LR is especially valuable for medicinal chemists because of its simplicity and the possibility of clear mechanistic interpretation.

Self-organizing neural integration of pose-motion features for human action recognition.

The visual recognition of complex, articulated human movements is fundamental for a wide range of artificial systems oriented toward human-robot communication, action classification, and action-driven perception. These challenging tasks may generally involve the processing of a huge amount of visual information and learning-based mechanisms for generalizing a set of training actions and classifying new samples. To operate in natural environments, a crucial property is the efficient and robust recognition of actions, also under noisy conditions caused by, for instance, systematic sensor errors and temporarily occluded persons. Studies of the mammalian visual system and its outperforming ability to process biological motion information suggest separate neural pathways for the distinct processing of pose and motion features at multiple levels and the subsequent integration of these visual cues for action perception. We present a neurobiologically-motivated approach to achieve noise-tolerant action recognition in real time. Our model consists of self-organizing Growing When Required (GWR) networks that obtain progressively generalized representations of sensory inputs and learn inherent spatio-temporal dependencies. During the training, the GWR networks dynamically change their topological structure to better match the input space. We first extract pose and motion features from video sequences and then cluster actions in terms of prototypical pose-motion trajectories. Multi-cue trajectories from matching action frames are subsequently combined to provide action dynamics in the joint feature space. Reported experiments show that our approach outperforms previous results on a dataset of full-body actions captured with a depth sensor, and ranks among the best results for a public benchmark of domestic daily actions.

Development of human B cells and antibodies following human hematopoietic stem cell transplantation to Rag2(-/-)γc(-/-) mice.

Humanized mice represent a valuable model system to study the development and functionality of the human immune system. In the RAG-hu mouse model highly immunodeficient Rag2(-/-)γc(-/-) mice are transplanted with human CD34(+) hematopoietic stem cells, resulting in human hematopoiesis and a predominant production of B and T lymphocytes. Human adaptive immune responses have been detected towards a variety of antigens in humanized mice but both cellular and humoral immune responses tend to be weak and sporadically detected. The underlying mechanisms for inconsistent responses are poorly understood. Here, we analyzed the kinetics of human B cell development and antibody production in RAG-hu mice to better understand the lack of effective antibody responses. We found that T cell levels in blood did not significantly change from 8 to 28 weeks post-engraftment, while B cells reached a peak at 14 weeks. Concentrations of 3 antibody classes (IgM, IgG, IgA) were found to be at levels about 0.1% or less of normal human levels, but human antibodies were still detected up to 32 weeks after engraftment. Human IgM was detected in 92.5% of animals while IgG and IgA were detected in about half of animals. We performed flow cytometric analysis of human B cells in bone marrow, spleen, and blood to examine the presence of precursor B cells, immature B cells, naïve B cells, and plasma B cells. We detected high levels of surface IgM(+) B cells (immature and naïve B cells) and low levels of plasma B cells in these organs, suggesting that B cells do not mature properly in this model. Low levels of human T cells in the spleen were observed, and we suggest that the lack of T cell help may explain poor B cell development and antibody responses. We conclude that human B cells that develop in humanized mice do not receive the signals necessary to undergo class-switching or to secrete antibody effectively, and we discuss strategies to potentially overcome these barriers.

Salud sexual y reproductiva en estudiantes universitarios de una institución de educación superior en Colombia / Sexual and reproductive health in university students at an institution of higher learning in Colombia

Objetivo Investigar en una universidad estatal del Caribe colombianoel estado de la salud sexual y reproductiva especialmente conocimientos sobre Enfermedades de Transmisión Sexual (ETS), fertilidad, sexualidad, embarazo y violencia. Métodos Estudio transversal tipo encuesta. Se seleccionaron estudiantes matriculados en el segundo período de 2010 que llenaron voluntariamente la encuesta. Los datos cualitativos fueron tabulados y graficados usando medidas de tendencia central para variables cuantitativas. Resultados La población tenía alrededor de 20 años de edad, provenía de la zona urbana (57,9%; IC95%=54,7-61,1), predominantemente heterosexual (89,7%) con edad de inicio de vida sexual antes de los 18 años, con 11,8 % de promiscuidad que usaba principalmente el preservativo como Método de Planificación Familiar (MPF) (55 %). Aunque poseían información previa sobre salud sexual, MPF y ETS, no se comportaban acorde con ello, por la desinformación sobre formas de transmisión del VIH, bajo nivel de realización de pruebas serológicas para ETS, así como conductas de riesgo (sexo/alcohol/drogas). El 12,3% tuvo antecedente de embarazo, violencia física (21,6%) y sexual (4,6%) con un predominante silencio de las víctimas de abuso sexual (61,8%). Conclusiones La muestra refleja la población universitaria de esta región del país. Se planea organizar un programa con apoyo médico y psicológico que permita disminuir los índices de ETS, embarazos no planificados, preparando al adolescente en esta importante etapa de su vida, sirviendo de modelo para otras universidades latinoamericanas.

Objective To investigate the state of sexual and reproductive health in students at a public university in the Colombian Caribbean, with an emphasis on sexually transmitted diseases (STDs), fertility, sexuality, pregnancy and violence. Methods Cross-sectional survey study. University students, enrolled in the second semester of 2010 and who completed a self-administered survey based on the Reproductive Health survey of the Pan American Health Organization, were selected. Qualitative data was tabulated and graphed using measures of central tendency for quantitative variables. Results The age of population studied was around 20 years old, came from the urban area (57.9 %; IC95 %=54.7-61.1), was predominantly heterosexual (89.7 %), with an age of initiation of sexual activity of less than 18 years old, 11.8 % promiscuity, mainly using the condom as a Family Planning Method (FPM) (55 %). Although they had prior information on sexual health, STDs and FPMs, they did not behave according to this due to low education about HIV transmission routes, low incidence of serological tests for STDs, and high risk behavior (sex/alcohol/drugs). It was observed that 12.3 % had a history of pregnancy, physical violence (21.6 %) and sexual violence (4.6 %) with a predominant silence from the victims of sexual abuse (61.8 %). Conclusion The sample reflects the student population in this region of Colombia. We plan to organize a health program with medical and psychological support to reduce the rates of STDs and unplanned pregnancies, preparing the adolescent for this important step in their life and serving as a model for other Latin American universities.

[Sexual and reproductive health in university students at an institution of higher learning in Colombia]. / Salud sexual y reproductiva en estudiantes universitarios de una institución de educación superior en Colombia.

OBJECTIVE: To investigate the state of sexual and reproductive health in students at a public university in the Colombian Caribbean, with an emphasis on sexually transmitted diseases (STDs), fertility, sexuality, pregnancy and violence. METHODS: Cross-sectional survey study. University students, enrolled in the second semester of 2010 and who completed a self-administered survey based on the Reproductive Health survey of the Pan American Health Organization, were selected. Qualitative data was tabulated and graphed using measures of central tendency for quantitative variables. RESULTS: The age of population studied was around 20 years old, came from the urban area (57.9 %; IC95 %=54.7-61.1), was predominantly heterosexual (89.7 %), with an age of initiation of sexual activity of less than 18 years old, 11.8 % promiscuity, mainly using the condom as a Family Planning Method (FPM) (55 %). Although they had prior information on sexual health, STDs and FPMs, they did not behave according to this due to low education about HIV transmission routes, low incidence of serological tests for STDs, and high risk behavior (sex/alcohol/drugs). It was observed that 12.3 % had a history of pregnancy, physical violence (21.6 %) and sexual violence (4.6 %) with a predominant silence from the victims of sexual abuse (61.8 %). CONCLUSION: The sample reflects the student population in this region of Colombia. We plan to organize a health program with medical and psychological support to reduce the rates of STDs and unplanned pregnancies, preparing the adolescent for this important step in their life and serving as a model for other Latin American universities.

Production and characterization of humanized Rag2-/-γc -/- mice.

Mice reconstituted with human immune cells represent a model to study the development and functionality of the human immune system. Recent improvements in humanized mice have resulted in multi-lineage hematopoiesis, prolonged human cell engraftment that is detectable in many mouse organs, and the ability to generate de novo human innate and adaptive immune responses. Here, we describe the methods used to produce and characterize humanized Rag2(-/-)γc(-/-) mice.

Valoración eco–sistémica de la naturaleza: hacia una ética emergente en los sistemas socio ecológicos adaptativos / Eco–systemic Assessment: towards an emergent ethics on the socioecologic adaptive systems

Con base en el modelo de la Evaluación de los Ecosistemas del Milenio, que liga la biodiversidad, los servicios eco–sistémicos, el bienestar humano y la libertad de elección, se proponen niveles y fuerzas opuestas de equilibrio que como procesos de auto–organización sustentan un nuevo sistema de valoración de la naturaleza. Las propiedades estabilizadoras y desestabilizadoras en cada nivel, se presentan además como atributos complejos del bien común ambiental, y propiedades emergentes y adaptativas en el propuesto sistema de valoración social de la naturaleza. El conjunto sugiere la emergencia de una nueva normativa de acción socio–ecológica para enfrentar el cambio ambiental global, que a su vez estaría sustentando una ética eco–sistémica.

Based upon the Millennium Ecosystem Assessment a model which link biodiversity, ecosystem services, human well-being and liberty of choice, levels and opposing self-organization forces are proposed, for a new valuation system of nature. The stabilizing and destabilizing properties in each level area presented as complex attributes or the environmental common good and adaptive emerging properties in the proposed system of valuation of nature. Overall, the emergence of a new action-driven and socio-ecological normative which would help society face up to global environmental changes, and sustain a new eco–systemic ethic.

Local delivery of sirolimus nanoparticles for the treatment of in-stent restenosis.

OBJECTIVES: To test the local delivery of sirolimus nanoparticles following percutaneous transluminal coronary angioplasty (PTCA) to treat in-stent restenosis (ISR) in a swine model. BACKGROUND: Coronary bare-metal stent (BMS) implantation reduces major adverse cardiac events when compared with PTCA; however, ISR rates remain high. METHODS: Eighteen swine underwent BMS deployment guided by intravascular ultrasound (IVUS). Of these, 16 developed ISR (1 stent/swine) and underwent angioplasty with a noncompliant balloon (PTCA-NC). The animals were then randomized into four groups for local infusion of sirolimus nanoparticles through a porous balloon catheter, as follows: (1) PTCA-NC alone (control); (2) PTCA-NC + (polylactic acid)-based nanoparticle formulation (anionic 1); (3) PTCA-NC + (polylactic-co-glycolic acid)-based nanoparticle formulation (anionic 2); and (4) PTCA-NC + Eudragit RS nanoparticle formulation (cationic). Coronary angiography and IVUS follow-up were performed 28 days after ISR treatment. RESULTS: There was one episode of acute coronary occlusion with the cationic formulation. Late area loss was similar in all groups at 28 days according to IVUS. However, luminal volume loss (control = 20.7%, anionic 1 = 4.0%, anionic 2 = 6.7%, cationic = 9.6%; P = 0.01) and neointimal volume gain (control = 68.7%, anionic 1 = 17.4%, anionic 2 = 29.5%, cationic = 31.2%; P = 0.019) were significantly reduced in all treatment groups, especially in anionic 1. CONCLUSIONS: PTCA-NC followed by local infusion of sirolimus nanoparticles was safe and efficacious to reduce neointima in this model, and this strategy may be a promising treatment for BMS ISR. Further studies are required to validate this method in humans.