Variational Nested Dropout

Nested dropout  is a variant of dropout operation that is able to order network parameters or features based on the pre-defined importance during training. It has been explored for: I. Constructing nested nets [11], [10]: the nested nets are neural networks whose architectures can be adjusted instantly during testing time, e.g., based on computational constraints. The nested dropout implicitly ranks the network parameters, generating a set of sub-networks such that any smaller sub-network forms the basis of a larger one. II. Learning ordered representation [48]: the nested dropout applied to the latent representation of a generative model (e.g., auto-encoder) ranks the features, enforcing explicit order of the dense representation over dimensions. However, the dropout rate is fixed as a hyper-parameter during the whole training process. For nested nets, when network parameters are removed, the performance decays in a human-specified trajectory rather than in a trajectory learned from data. For generative models, the importance of features is specified as a constant vector, restraining the flexibility of representation learning. To address the problem, we focus on the probabilistic counterpart of the nested dropout. We propose a variational nested dropout (VND) operation that draws samples of multi-dimensional ordered masks at a low cost, providing useful gradients to the parameters of nested dropout. Based on this approach, we design a Bayesian nested neural network that learns the order knowledge of the parameter distributions. We further exploit the VND under different generative models for learning ordered latent distributions. In experiments, we show that the proposed approach outperforms the nested network in terms of accuracy, calibration, and out-of-domain detection in classification tasks. It also outperforms the related generative models on data generation tasks.

Accelerating Monte Carlo Bayesian Prediction via Approximating Predictive Uncertainty Over the Simplex

Estimating the predictive uncertainty of a Bayesian learning model is critical in various decision-making problems, e.g., reinforcement learning, detecting the adversarial attack, self-driving car. As the model posterior is almost always intractable, most efforts were made on finding an accurate approximation to the true posterior. Even though a decent estimation of the model posterior is obtained, another approximation is required to compute the predictive distribution over the desired output. A common accurate solution is to use Monte Carlo (MC) integration. However, it needs to maintain a large number of samples, and evaluate the model repeatedly, and average multiple model outputs. In many real-world cases, this is computationally prohibitive. In this work, assuming that the exact posterior or a decent approximation is obtained, we propose a generic framework to approximate the output probability distribution induced by the model posterior with a parameterized model and in an amortized fashion. The aim is to approximate the predictive uncertainty of a specific Bayesian model, meanwhile alleviating the heavy workload of MC integration at testing time. The proposed method is universally applicable to Bayesian classification models that allow for posterior sampling. Theoretically, we show that the idea of amortization incurs no additional costs on approximation performance. Empirical results validate the strong practical performance of our approach.

Faster super-resolution imaging of high density molecules via a cascading algorithm based on compressed sensing

This paper proposes a cascading algorithm (CSR) based on compressed sensing, which aims to reduce intensive computations in super-resolution imaging of fluorescence microscopy. Performance of existing algorithms such as CVX and L1H drop sharply when applied to obtain finer images with high density molecules. CSR fully exploits the extreme sparsity property of molecules in the compressed sensing model and progressively restricts solution space stage by stage. We perform a comprehensive study of existing algorithms and the proposed algorithm under different resolutions and molecules' densities. Simulation and experimental results confirm the performance advantage of CSR when applied to recover dense molecules.

Dynamic localisation of mature microRNAs in Human nucleoli is influenced by exogenous genetic materials

Although microRNAs are commonly known to function as a component of RNA-induced silencing complexes in the cytoplasm, they have been detected in other organelles, notably the nucleus and the nucleolus, of mammalian cells. We have conducted a systematic search for miRNAs in HeLa cell nucleoli, and identified 11 abundant miRNAs with a high level of nucleolar accumulation. Through in situ hybridisation, we have localised these miRNAs, including miR-191 and miR-484, in the nucleolus of a diversity of human and rodent cell lines. The nucleolar association of these miRNAs is resistant to various cellular stresses, but highly sensitive to the presence of exogenous nucleic acids. Introduction of both single- and double-stranded DNA as well as double stranded RNA rapidly induce the redistribution of nucleolar miRNAs to the cytoplasm. A similar change in subcellular distribution is also observed in cells infected with the influenza A virus. The partition of miRNAs between the nucleolus and the cytoplasm is affected by Leptomycin B, suggesting a role of Exportin-1 in the intracellular shuttling of miRNAs. This study reveals a previously unknown aspect of miRNA biology, and suggests a possible link between these small noncoding RNAs and the cellular management of foreign genetic materials.

Amphetamine increases glutamate efflux in the rat ventral tegmental area by a mechanism involving glutamate transporters and reactive oxygen species

We have shown that amphetamine produces a delayed and sustained increase in glutamate levels in the ventral tegmental area, a region containing dopamine cell bodies important in acute and chronic effects of amphetamine administration. The present study characterized the mechanism underlying amphetamine-induced glutamate efflux. It was abolished by the glutamate uptake inhibitor dihydrokainate, but unaffected by perfusion with a low Ca(2+)/high Mg(2+) solution, implicating glutamate transporters. Because reactive oxygen species inhibit glutamate uptake, we examined the effect of amphetamine on hydroxyl radical formation by perfusing with D-phenylalanine (5 mM) and monitoring p-tyrosine production. Although no increase in hydroxyl radical formation was detected, D-phenylalanine completely prevented the amphetamine-induced increase in glutamate efflux, as did systemic injection of another trapping agent, alpha-phenyl-N-tert-butyl nitrone (60 mg/kg). Thus, amphetamine-induced glutamate efflux may involve reactive oxygen species. In other studies, we found that repeated coadministration of alpha-phenyl-N-tert-butyl nitrone with amphetamine attenuated the development of behavioral sensitization. This supports prior results indicating that the increase in glutamate efflux produced by each amphetamine injection in a chronic regimen is important in triggering drug-induced adaptations in ventral tegmental area dopamine neurons, and that such adaptations may in part represent a response to metabolic and oxidative stress

Environmental modulation of the response to amphetamine: dissociation between changes in behavior and changes in dopamine and glutamate overflow in the rat striatal complex

RATIONALE

We have previously shown that environmental novelty enhances the behavioral activating effects of amphetamine and amphetamine-induced expression of the immediate early gene c-fos in the striatal complex, particularly in the most caudal portion of the caudate. In contrast, we found no effect of novelty on the ability of amphetamine to induce dopamine (DA) overflow in the rostral caudate or in the core of the nucleus accumbens.

OBJECTIVES

The twofold aim of the present study was to determine the effect of environmental novelty on (1) amphetamine-induced DA overflow in the shell of the nucleus accumbens and in the caudal portions of the caudate, and (2) glutamate and aspartate overflow in the caudal portions of the caudate.

METHODS

Two groups of rats with a unilateral 6-hydroxydopamine lesion of the mesostriatal dopaminergic system received amphetamine (0.5 mg/kg, i.v.) in physically identical cages. For one group, the cages were also the home environment, whereas, for the other group, they were a completely novel environment. In vivo microdialysis was used to estimate DA, glutamate, and aspartate concentrations.

RESULTS

Environmental novelty enhanced amphetamine-induced rotational behavior (experiments 1-3) but did not alter amphetamine-induced DA overflow in either the shell of the nucleus accumbens (experiment 1) or the caudate (experiment 2). In addition, the ability of environmental novelty to enhance amphetamine-induced behavioral activation was not associated with changes in glutamate or aspartate efflux in the caudate (experiment 3).

CONCLUSIONS

The present data indicate that the psychomotor activating effects of amphetamine can be modulated by environmental context independent of its primary neuropharmacological actions in the striatal complex.

Amphetamine-induced glutamate efflux in the rat ventral tegmental area is prevented by MK-801, SCH 23390, and ibotenic acid lesions of the prefrontal cortex

We showed previously that amphetamine challenge produces a delayed increase in glutamate efflux in the ventral tegmental area of both naive and chronic amphetamine-treated rats. The present study examined the mechanisms underlying this response. The NMDA receptor antagonist MK-801 (0.1 mg/kg, i.p.) or the D1 dopamine receptor antagonist SCH 23390 (0.1 mg/kg, i.p.), given 30 min before acute amphetamine (5 mg/kg, i.p.), prevented amphetamine-induced glutamate efflux. Neither antagonist by itself altered glutamate efflux. Ibotenic acid lesions of the prefrontal cortex similarly prevented amphetamine-induced glutamate efflux, while producing a trend toward decreased basal glutamate levels (82.8% of sham group). Previous work has shown that the doses of NMDA and D1 receptor antagonists used in this study prevent the induction of behavioral sensitization when coadministered repeatedly with amphetamine, and that identical prefrontal cortex lesions performed before repeated amphetamine prevent the induction of ambulatory sensitization. Thus, treatments that prevent acute amphetamine from elevating glutamate efflux in the ventral tegmental area also prevent repeated amphetamine from eliciting behavioral sensitization. These findings suggest that repeated elevation of glutamate levels during a chronic amphetamine regimen may contribute to the cascade of neuroadaptations within the ventral tegmental area that enables the induction of sensitization.

Amphetamine and D1 dopamine receptor agonists produce biphasic effects on glutamate efflux in rat ventral tegmental area: modification by repeated amphetamine administration

Amphetamine or selective D1 and D2 dopamine receptor agonists and antagonists were administered to the ventral tegmental area (VTA) through a microdialysis probe to determine their effects on glutamate and aspartate efflux in rats pretreated for 5 days with vehicle or 5 mg/kg (+)-amphetamine sulfate. In vehicle rats, glutamate efflux declined during 2 h of perfusion with the D1 receptor agonist SKF-82958 (10 and 100 microM). After SKF-82958 perfusion ended, glutamate efflux rebounded to basal levels and continued to increase gradually over the next 2 h. A similar biphasic pattern was observed with intra-VTA amphetamine (10 and 100 microM) and with another D1 agonist (100 microM SKF-38393). The biphasic effects of SKF-82958 were prevented by coperfusion with a D1 antagonist (SCH-23390; 30 microM). Glutamate efflux was unaffected by a D2 agonist (100 microM quinpirole) and by D1 or D2 antagonists administered alone (SCH 23390 and eticlopride; 30 microM). In amphetamine-pretreated rats tested 2 days after the last injection, both the decrease during SKF-82958 perfusion and the delayed increase in glutamate efflux were attenuated. In rats tested 12-14 days after the last amphetamine injection, only the decrease during SKF-82958 perfusion was attenuated. None of these drug treatments produced consistent effects on aspartate efflux. We showed previously that systemic amphetamine (5 mg/kg, i.p.) has no immediate effect on VTA glutamate efflux but produces a delayed increase in glutamate efflux that reaches statistical significance 2-3 h after injection. Because behavioral sensitization can be elicited either by repeated systemic or repeated intra-VTA administration, neurochemical effects common to both routes (such as the delayed increase in glutamate efflux) are most likely to contribute to its induction.

Effects of the AMPA receptor antagonist NBQX on the development and expression of behavioral sensitization to cocaine and amphetamine

We examined the effect of 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo(f)quinoxaline (NBQX), an antagonist of the alpha-amino-3-hydroxy-5-methyl-4-isoxazole propionate (AMPA) subtype of glutamate receptor, on the development and expression of behavioral sensitization to amphetamine and cocaine in rats. A single injection of NBQX (12.5 mg/kg) administered 30 min prior to cocaine during the induction phase (days 1-5) prevented the development of cocaine sensitization, assessed by responsiveness to cocaine challenge on day 8. This NBQX regimen did not affect development of amphetamine sensitization. Two pretreatment injections of NBQX, one 20 min before and one 70 min after amphetamine on each day of the induction phase (days 1-6), did not affect sensitization of stereotypy but prevented sensitization of post-stereotypy ambulatory hyperactivity (both assessed by responsiveness to amphetamine challenge on day 8). The effect of NBQX on ambulatory sensitization was dose-dependent (attenuation with 12.5 mg/kg, complete prevention with 25 mg/kg). In contrast to its effects on development, NBQX (25 mg/kg) did not prevent expression of sensitization to cocaine or amphetamine. NBQX itself exerted no significant effects on locomotor activity in either drug-naive rats or rats that had received either NBQX or amphetamine repeatedly. These findings support a requirement for AMPA receptor stimulation in the development of locomotor sensitization to cocaine and amphetamine, but suggest a different mechanism for sensitization of amphetamine stereotypy.

Repeated amphetamine administration alters the expression of mRNA for AMPA receptor subunits in rat nucleus accumbens and prefrontal cortex

Recent evidence suggests that behavioral sensitization to amphetamine is associated with alterations in excitatory amino acid (EAA) transmission in perikarya (ventral tegmental area) and terminal regions (nucleus accumbens [NAc]) of the mesoaccumbens dopamine system. The present study determined whether repeated amphetamine administration alters expression of mRNAs for AMPA receptor subunits. We studied the NAc, because it is the site of expression of amphetamine sensitization, and the prefrontal cortex (PFC), because it is the origin of EAA projections that regulate the mesoaccumbens dopamine system. Rats were treated for 5 days with 5 mg/kg/day amphetamine sulfate or vehicle (controls) and perfused 3 or 14 days after the last injection. We used a novel in situ hybridization method that allows quantification of mRNA levels [Lu et al. (1996) J. Neurosci. Methods, 65:69-76]. Repeated amphetamine administration decreased levels of GluR1 and GluR2 but not GluR3 mRNAs in both core and shell subregions of the NAc at the 14 day withdrawal time; no changes were observed after 3 days of withdrawal. In contrast, levels of GluR1 mRNA in the PFC were increased at 3 but not 14 days of withdrawal, while GluR2 and 3 mRNAs were unchanged. Levels of GluR4 mRNA were very low in both NAc and PFC. Functional properties of heteromeric AMPA receptors are determined by subunit composition. Thus, the observed changes in mRNAs for AMPA receptor subunits may result in altered AMPA transmission in NAc and PFC. This, in turn, may influence the responsiveness of the mesoaccumbens DA system to psychomotor stimulants and potentially contribute to behavioral sensitization.

Acute and repeated systemic amphetamine administration: effects on extracellular glutamate, aspartate, and serine levels in rat ventral tegmental area and nucleus accumbens

Recent work indicates an important role for excitatory amino acids in behavioral sensitization to amphetamine. We therefore examined, using in vivo microdialysis in awake rats, the effects of amphetamine on efflux of glutamate, aspartate, and serine in the ventral tegmental area and nucleus accumbens, brain regions important for the initiation and expression of amphetamine sensitization, respectively. Water-pretreated and amphetamine-pretreated rats were compared to determine if sensitization altered such effects. In both brain regions, Ca2+-dependent efflux of glutamate accounted for approximately 20% of basal glutamate efflux. A challenge injection of water or 2.5 mg/kg of amphetamine did not significantly alter glutamate, aspartate, or serine efflux in the ventral tegmental area or nucleus accumbens of water- or amphetamine-pretreated rats. However, 5 mg/kg of amphetamine produced a gradual increase in glutamate efflux in both regions that did not reverse, was observed in both water- and amphetamine-pretreated rats, and was prevented by haloperidol. Although increased glutamate efflux occurred with too great a delay to mediate acute behavioral responses to amphetamine, it is possible that repeated augmentation of glutamate efflux during repeated amphetamine administration results in compensatory changes in levels of excitatory amino acid receptors in the ventral tegmental area and nucleus accumbens that contribute to development of expression of amphetamine sensitization.

Effects of lesions of prefrontal cortex, amygdala, or fornix on behavioral sensitization to amphetamine: comparison with N-methyl-D-aspartate antagonists

Behavioral sensitization to amphetamine involves the mesoaccumbens dopamine system and is accompanied by cellular changes in this system. Excitatory amino acid antagonists, when co-administered with amphetamine, prevent both behavioral sensitization and associated changes in the mesoaccumbens dopamine system. This suggests that excitatory amino acid-dependent events are critical to the initiation of sensitization. This study sought to identify excitatory amino acid projections required for sensitization, focusing on projections to the nucleus accumbens or ventral tegmental area. The major excitatory projections to the nucleus accumbens originate in the prefrontal cortex, amygdala and hippocampus. The prefrontal cortex and amygdala also send excitatory projections to the ventral tegmental area. Ibotenic acid lesions of the prefrontal cortex or amygdala and electrolytic lesions of the fornix were performed in rats. After one week of recovery, rats were treated with water or 2.5 mg/kg amphetamine for six days and challenged with amphetamine on day 8. Activity was tested in photobeam cages on days 1 and 8. On day 1, control and sham-lesioned rats exhibited stereotyped behaviors followed by a period of post-stereotypy locomotion. On day 8, sensitization was evident as an enhancement of both stereotypy and post-stereotypy locomotion. Co-administration of N-methyl-D-aspartate antagonists [MK-801 (dizocilpine maleate) or CGS 19755] with amphetamine prevented the development of sensitization of both stereotypy and post-stereotypy locomotion. Neither antagonist, however, prevented the expression of sensitization. None of the lesions completely mimicked these effects of N-methyl-D-aspartate antagonists. Lesions of hippocampal projections traveling in the fornix produced a general disinhibition of locomotor activity, but did not prevent sensitization of either stereotypy or post-stereotypy locomotion. Lesions of the prefrontal cortex failed to prevent sensitization of stereotypy was obtained following repeated amphetamine administration. However, like prefrontal cortical lesions, amygdala lesions prevented sensitization of post-stereotypy locomotion. When interpreted in the light of previous studies demonstrating the importance of the ventral tegmental area in the initiation of sensitization, the present results suggest a likely role for neuronal circuits involving the prefrontal cortex, amygdala and ventral tegmental area in the development of sensitization of post-stereotypy locomotion following repeated amphetamine administration. Such circuits may initiate sensitization through a mechanism involving excitatory amino acid regulation of the activity of mesoaccumbens dopamine neurons. Parallel circuits, involving other brain regions, may similarly contribute to sensitization of stereotyped behaviors.

MK-801 does not prevent acute stimulatory effects of amphetamine or cocaine on locomotor activity or extracellular dopamine levels in rat nucleus accumbens

Recent work has shown that the development of behavioral sensitization to cocaine, amphetamine, and morphine is prevented by coadministration of N-methyl-D-aspartate (NMDA) antagonists such as MK-801. This suggests that NMDA receptors mediate long-term changes in neuronal responsiveness essential for the development of behavioral sensitization, similar to their role in other forms of neuronal plasticity. However, other studies, suggesting that NMDA receptor antagonists interfere with acute behavioral effects of psychomotor stimulants, call this interpretation into question and suggest that the ability of NMDA antagonists to prevent sensitization may reflect blockade of the acute effects of psychomotor stimulants. To examine this issue, behavioral and microdialysis studies assessed the effect of pretreatment with 0.1 mg/kg MK-801 on the ability of amphetamine and cocaine to stimulate locomotor activity and elevate extracellular dopamine (DA) levels in nucleus accumbens; this dose of MK-801 prevents sensitization when coadministered repeatedly with these stimulants. MK-801 pretreatment enhanced amphetamine-stimulated horizontal locomotion and stereotyped behavior. MK-801 pretreatment produced a modest attenuation of cocaine-stimulated horizontal locomotion, which may have reflected enhancement by MK-801 of certain components of cocaine-stimulated stereotypy. There was no effect of MK-801 pretreatment on the ability of amphetamine or cocaine to elevate extracellular DA levels in nucleus accumbens. These results suggest that the acute effects of cocaine and amphetamine on locomotor activity and extracellular DA levels are not prevented by MK-801, and that MK-801 must act through other mechanisms to prevent the development of behavioral sensitization.