arXiv subjects

Tatsuya Konishi

Publications and source records attributed to Tatsuya Konishi.

14 recordsLinked to original sources

arXiv2026

An Exposure-Time-Aligned Primary-Path Architecture for Autonomous-Driving ECUs

While end-to-end (E2E) autonomous driving has become the dominant research direction, production vehicles continue to rely on modular multi-NN pipelines for a non-trivial transitional period. The subject of this paper is the design of an architecture that, during this phase, supports a modular pipeline and an E2E path side by side and embeds a path for staged migration. Transplanted to a production SoC, egalitarian late fusion is compute-inefficient and offers no natural unit for staged E2E substitution. As an alternative, we propose three design principles: (i) Primary-Path, which selects a primary perception-to-planner chain and prioritizes its enclosure within a single SoC over the non-critical paths, (ii) Exposure-Time-Aligned, which propagates the primary sensor's exposure time $τ_{\mathrm{exp}}$ as a tag along the chain and event-drives the fusion node on matched $τ_{\mathrm{exp}}$ rather than a fixed cycle, and (iii) Co-Path Coexistence, which, building on (i) and (ii), lets an E2E output path co-run with the modular pipeline within the same $τ_{\mathrm{exp}}$ cycle. On a Dual-SoC production AD-ECU, the implementation achieves a camera-shutter to planner-output latency of 296 ms on average within a 350 ms design budget, with the modular pipeline and the E2E path coexisting within the shared $τ_{\mathrm{exp}}$ cycle.

cs.RO↗

arXiv2025

CADE: Continual Weakly-supervised Video Anomaly Detection with Ensembles

Video anomaly detection (VAD) has long been studied as a crucial problem in public security and crime prevention. In recent years, weakly-supervised VAD (WVAD) have attracted considerable attention due to their easy annotation process and promising research results. While existing WVAD methods tackle mainly on static datasets, the possibility that the domain of data can vary has been neglected. To adapt such domain-shift, the continual learning (CL) perspective is required because otherwise additional training only with new coming data could easily cause performance degradation for previous data, i.e., forgetting. Therefore, we propose a brand-new approach, called Continual Anomaly Detection with Ensembles (CADE) that is the first work combining CL and WVAD viewpoints. Specifically, CADE uses the Dual-Generator(DG) to address data imbalance and label uncertainty in WVAD. We also found that forgetting exacerbates the "incompleteness'' where the model becomes biased towards certain anomaly modes, leading to missed detections of various anomalies. To address this, we propose to ensemble Multi-Discriminator (MD) that capture missed anomalies in past scenes due to forgetting, using multiple models. Extensive experiments show that CADE significantly outperforms existing VAD methods on the common multi-scene VAD datasets, such as ShanghaiTech and Charlotte Anomaly datasets.

cs.CV↗

arXiv2025

Learning After Model Deployment

In classic supervised learning, once a model is deployed in an application, it is fixed. No updates will be made to it during the application. This is inappropriate for many dynamic and open environments, where unexpected samples from unseen classes may appear. In such an environment, the model should be able to detect these novel samples from unseen classes and learn them after they are labeled. We call this paradigm Autonomous Learning after Model Deployment (ALMD). The learning here is continuous and involves no human engineers. Labeling in this scenario is performed by human co-workers or other knowledgeable agents, which is similar to what humans do when they encounter an unfamiliar object and ask another person for its name. In ALMD, the detection of novel samples is dynamic and differs from traditional out-of-distribution (OOD) detection in that the set of in-distribution (ID) classes expands as new classes are learned during application, whereas ID classes is fixed in traditional OOD detection. Learning is also different from classic supervised learning because in ALMD, we learn the encountered new classes immediately and incrementally. It is difficult to retrain the model from scratch using all the past data from the ID classes and the novel samples from newly discovered classes, as this would be resource- and time-consuming. Apart from these two challenges, ALMD faces the data scarcity issue because instances of new classes often appear sporadically in real-life applications. To address these issues, we propose a novel method, PLDA, which performs dynamic OOD detection and incremental learning of new classes on the fly. Empirical evaluations will demonstrate the effectiveness of PLDA.

cs.LG↗

arXiv2024

Open-World Continual Learning: Unifying Novelty Detection and Continual Learning

As AI agents are increasingly used in the real open world with unknowns or novelties, they need the ability to (1) recognize objects that (a) they have learned before and (b) detect items that they have never seen or learned, and (2) learn the new items incrementally to become more and more knowledgeable and powerful. (1) is called novelty detection or out-of-distribution (OOD) detection and (2) is called class incremental learning (CIL), which is a setting of continual learning (CL). In existing research, OOD detection and CIL are regarded as two completely different problems. This paper first provides a theoretical proof that good OOD detection for each task within the set of learned tasks (called closed-world OOD detection) is necessary for successful CIL. We show this by decomposing CIL into two sub-problems: within-task prediction (WP) and task-id prediction (TP), and proving that TP is correlated with closed-world OOD detection. The key theoretical result is that regardless of whether WP and OOD detection (or TP) are defined explicitly or implicitly by a CIL algorithm, good WP and good closed-world OOD detection are necessary and sufficient conditions for good CIL, which unifies novelty or OOD detection and continual learning (CIL, in particular). We call this traditional CIL the closed-world CIL as it does not detect future OOD data in the open world. The paper then proves that the theory can be generalized or extended to open-world CIL, which is the proposed open-world continual learning, that can perform CIL in the open world and detect future or open-world OOD data. Based on the theoretical results, new CIL methods are also designed, which outperform strong baselines in CIL accuracy and in continual OOD detection by a large margin.

cs.LG↗

arXiv2023

Parameter-Level Soft-Masking for Continual Learning

Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Although several techniques have achieved learning with no CF, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction in capacity usage. To our knowledge, this is the first work that soft-masks a model at the parameter-level for continual learning. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while mitigating CF.

cs.LG↗

arXiv2023

Learnability and Algorithm for Continual Learning

This paper studies the challenging continual learning (CL) setting of Class Incremental Learning (CIL). CIL learns a sequence of tasks consisting of disjoint sets of concepts or classes. At any time, a single model is built that can be applied to predict/classify test instances of any classes learned thus far without providing any task related information for each test instance. Although many techniques have been proposed for CIL, they are mostly empirical. It has been shown recently that a strong CIL system needs a strong within-task prediction (WP) and a strong out-of-distribution (OOD) detection for each task. However, it is still not known whether CIL is actually learnable. This paper shows that CIL is learnable. Based on the theory, a new CIL algorithm is also proposed. Experimental results demonstrate its effectiveness.

cs.LG↗

arXiv2023

Continual Pre-training of Language Models

Language models (LMs) have been instrumental for the rapid advance of natural language processing. This paper studies continual pre-training of LMs, in particular, continual domain-adaptive pre-training (or continual DAP-training). Existing research has shown that further pre-training an LM using a domain corpus to adapt the LM to the domain can improve the end-task performance in the domain. This paper proposes a novel method to continually DAP-train an LM with a sequence of unlabeled domain corpora to adapt the LM to these domains to improve their end-task performances. The key novelty of our method is a soft-masking mechanism that directly controls the update to the LM. A novel proxy is also proposed to preserve the general knowledge in the original LM. Additionally, it contrasts the representations of the previously learned domain knowledge (including the general knowledge in the pre-trained LM) and the knowledge from the current full network to achieve knowledge integration. The method not only overcomes catastrophic forgetting, but also achieves knowledge transfer to improve end-task performances. Empirical evaluation demonstrates the effectiveness of the proposed method.

cs.CL↗

arXiv2022

A Theoretical Study on Solving Continual Learning

Continual learning (CL) learns a sequence of tasks incrementally. There are two popular CL settings, class incremental learning (CIL) and task incremental learning (TIL). A major challenge of CL is catastrophic forgetting (CF). While a number of techniques are already available to effectively overcome CF for TIL, CIL remains to be highly challenging. So far, little theoretical study has been done to provide a principled guidance on how to solve the CIL problem. This paper performs such a study. It first shows that probabilistically, the CIL problem can be decomposed into two sub-problems: Within-task Prediction (WP) and Task-id Prediction (TP). It further proves that TP is correlated with out-of-distribution (OOD) detection, which connects CIL and OOD detection. The key conclusion of this study is that regardless of whether WP and TP or OOD detection are defined explicitly or implicitly by a CIL algorithm, good WP and good TP or OOD detection are necessary and sufficient for good CIL performances. Additionally, TIL is simply WP. Based on the theoretical result, new CIL methods are also designed, which outperform strong baselines in both CIL and TIL settings by a large margin.

cs.LG↗

arXiv2019

Using simulated annealing for locating array construction

Context: Combinatorial interaction testing is known to be an efficient testing strategy for computing and information systems. Locating arrays are mathematical objects that are useful for this testing strategy, as they can be used as a test suite that enables fault localization as well as fault detection. In this application, each row of an array is used as an individual test. Objective: This paper proposes an algorithm for constructing locating arrays with a small number of rows. Testing cost increases as the number of tests increases; thus the problem of finding locating arrays of small sizes is of practical importance. Method: The proposed algorithm uses simulation annealing, a meta-heuristic algorithm, to find locating array of a given size. The whole algorithm repeatedly executes the simulated annealing algorithm by dynamically varying the input array size. Results: Experimental results show 1) that the proposed algorithm is able to construct locating arrays for problem instances of large sizes and 2) that, for problem instances for which nontrivial locating arrays are known, the algorithm is often able to generate locating arrays that are smaller than or at least equal to the known arrays. Conclusion: Based on the results, it is concluded that the proposed algorithm can produce small locating arrays and scale to practical problems.

cs.SE↗

arXiv2019

Finding minimum locating arrays using a CSP solver

Combinatorial interaction testing is an efficient software testing strategy. If all interactions among test parameters or factors needed to be covered, the size of a required test suite would be prohibitively large. In contrast, this strategy only requires covering $t$-wise interactions where $t$ is typically very small. As a result, it becomes possible to significantly reduce test suite size. Locating arrays aim to enhance the ability of combinatorial interaction testing. In particular, $(\overline{1}, t)$-locating arrays can not only execute all $t$-way interactions but also identify, if any, which of the interactions causes a failure. In spite of this useful property, there is only limited research either on how to generate locating arrays or on their minimum sizes. In this paper, we propose an approach to generating minimum locating arrays. In the approach, the problem of finding a locating array consisting of $N$ tests is represented as a Constraint Satisfaction Problem (CSP) instance, which is in turn solved by a modern CSP solver. The results of using the proposed approach reveal many $(\overline{1}, t)$-locating arrays that are smallest known so far. In addition, some of these arrays are proved to be minimum.

cs.SE↗

arXiv2010

Electric Conductance of Rh Atomic Contacts under Electrochemical Potential Control

The electric conductance of Rh atomic contacts was investigated under the electrochemical potential control. The conductance histogram of Rh atomic contacts varied with the electrochemical potential. When the electrochemical potential of the contact was kept at $Φ_{0}$= 0.1 V vs. Ag/AgCl (Rh potential), the conductance histogram did not show any features. At $Φ_{0}$= -0.1 V (under potential deposited hydrogen potential), the conductance histogram showed a feature around 2.3 $G_{0}$ ($G_{0}$ =2$e^{2}/h$), which agreed with the conductance value of a clean Rh atomic contact, which was observed in ultrahigh vacuum at low temperature. At $Φ_{0}$= -0.25 V (over potential deposited hydrogen potential), the conductance histogram showed features around 0.3 and 1.0 $G_{0}$. The conductance behavior of the Rh atomic contact was discussed by comparing previously reported results of other metals, Au, Ag, Cu, Pt, Pd, Ni, Co, and Fe. The conductance behavior of the metal atomic contacts related with the strength of the interaction between hydrogen and metal surface.

cond-mat.mes-hall↗

arXiv2008

Three reversible states controlled on a gold monoatomic contact by the electrochemical potential

Conductance of an Au mono atomic contact was investigated under the electrochemical potential control. The Au contact showed three different behaviors depending on the potential: 1 $G_{0}$ ($G_{0}$ = $2e^{2}/h$), 0.5 $G_{0}$ and not-well defined values below 1 $G_{0}$ were shown when the potential of the contact was kept at -0.6 V (double layer potential), -1.0 V (hydrogen evolution potential), and 0.8 V (oxide formation potential) versus Ag/AgCl in 0.1 M Na$_{2}$SO$_{4}$ solution, respectively. These three reversible states and their respective conductances could be fully controlled by the electrochemical potential. These changes in the conductance values are discussed based on the proposed structure models of hydrogen adsorbed and oxygen incorporated on an Au mono atomic contact.

cond-mat.mtrl-sci↗

arXiv2006

Conductance Bistability of Gold Nano-wire at Room Temperature

Quantized conductance behavior of gold nano wires was studied under electrochemical potential control. We fabricated 1 nm long mono atomic wires in solution at room temperature. Electrochemical potential significantly affected the stability of the mono atomic wire and fractional conductance peak occurrence in the conductance histogram. We revealed that the hydrogen adsorption on gold mono atomic wires was a decisive factor of the fractional peak, which was originated from the dynamic structural transition between two bistable states of the mono atomic wire showing the unit and the fractional values of the conductance. We could tune the stability of these bistable states to make the fractional conductance state preferable.

cond-mat.mtrl-sci↗

arXiv2005

Hydrogen-assisted stabilization of Ni nanowires in solution

We have studied conductance characteristics of mechanically fabricated Ni nanoconstrictions under controlling electrochemical potential and pH of the electrolyte. Conductance histogram showed clear feature peaked at 1-1.5 $G_{0}$ (=$2e^{2}/h$) when the potential of the constriction was kept at more negative potential than -900 mV vs. Ag/AgCl in pH=3.7. Comparable feature also appeared at more positive potential when lower pH solution was used. We have revealed that Ni mono atomic contact or mono atomic wire can be stabilized in solution at room temperature under the hydrogen evolution.

cond-mat.mtrl-sci↗